Gemini 4 Neon

(blog.google)

916 points by yigitdemirag17 hours ago

143 comments

  • HJain1310 hours ago
    Edited: For formatting and from feedback<p>So if I am getting this right?<p><pre><code> +-------------------------------+--------------+--------+--------+----------+-------+ | Lab | XL | L | M+ | M | S&#x2F;XS | +-------------------------------+--------------+--------+--------+----------+-------+ | Google (Gemini 4 Family #1) | | Argon | | | | | Google (Gemini 4 Family #2) | | Carbon | Neon | Hydrogen | | | Anthropic (Claude 5.2 Family) | Fable&#x2F;Mythos | | | | | | Anthropic (Claude 5.5 Family) | | Opus | | Sonnet | Haiku | | OpenAI (GPT 6.2 Family) | Astra | Sol | | | | | OpenAI (GPT 6.1 Family) | | | | Terra | | | OpenAI (GPT 6 Family) | | | | | Luna | +-------------------------------+--------------+--------+--------+----------+-------+ </code></pre> Notes:<p><pre><code> * Gemini 4 Family - Argon -- ++Train --&gt; Carbon - Carbon -- Distill --&gt; Neon, Hydrogen * Gemini 3 Family - 3 Pro -- ++Train --&gt; 3.1 Pro - 3 Pro -- Distill --&gt; 3 Flash - 3 Flash -- ++Train --&gt; 3.5 Flash ... -- ++Train --&gt; ... 3.8 Flash * &quot;S&#x2F;XS&quot; classification is due to weirdly low pricing for models by Ant&#x2F;OAI and not by raw perf</code></pre>
    • riskassessment9 hours ago
      I&#x27;m not sure it even makes sense to align models on price given that Neon appears to be considerably better than Sonnet&#x2F;Terra. Not to mention that there are rumors of a flash-lite type model even faster than Hydrogen.
      • locknitpicker8 hours ago
        &gt; I&#x27;m not sure it even makes sense to align models on price given that Neon appears to be considerably better than Sonnet&#x2F;Terra.<p>Models are nowadays compared in terms of their position with regards to a Pareto front assessing traits such as capabilities and cost because there are clear tradeoffs between cost and capabilities.<p>I&#x27;m sure we are going to see speed show up in the analysis in the near future.
      • HJain138 hours ago
        Yeah; I guess. Updated
      • mkagenius5 hours ago
        So total names to remember, hydrogen, argon, carbon, neon, luna, sol, terra, astra, opus, fable, sonnet, haiku<p>Then each one has a version. Then each one has an effort of thinking, low, mid, high, extra, ultra.<p>Good lord.
        • KPGv214 minutes ago
          Yeah, I just use whatever agy sets as the model when I fire up the IDE or CLI. I don&#x27;t hit the limit on my personal projects (I&#x27;m retired), and it&#x27;s fast and good every time. And free because I get some kind of bonkers usage caps with my Google Fi.<p>I know I need to change the hard-coded model name in my foreign language acquisition app because I guess with 4 being the &quot;mid-range&quot;, the 3.8 I use is going to be deprecated?
        • khafra3 hours ago
          [dead]
        • drob51818 minutes ago
          But wait! There’s more!
        • ozgung1 hour ago
          Yeah but all the names are rankable with the exception of fable.
      • trenchgun8 hours ago
        Yeah Neon competes with Opus 5.5
        • irthomasthomas3 hours ago
          Neon is within about 1% of Fable 5.2&#x27;s performance for about 40% the cost on vals.ai, so surely it beats opus 5.5 and competes with Fable 5.2?<a href="https:&#x2F;&#x2F;www.vals.ai&#x2F;benchmarks&#x2F;vals_index" rel="nofollow">https:&#x2F;&#x2F;www.vals.ai&#x2F;benchmarks&#x2F;vals_index</a>
    • seanmcdirmid7 hours ago
      I had a chat with Gemini this morning, Argon seems to be deprecated and replaced by Neon, so neon just occupies argon’s tier now. Argon is no longer selectable on my ultra configured antigravity account.
      • zeroq7 hours ago
        Carbon costs the same as Argon did. Neon is a cheaper model.<p>And after short session, I don&#x27;t think these models are comparable.<p>Feels like another rug pull.
        • kzrdude4 hours ago
          It&#x27;s a guest house where rugs are constantly changing, often while the guests are still dancing, it&#x27;s just part of the scenery at this point. Just remember to not stop dancing...
        • anentropic5 hours ago
          Currently Neon is more expensive on OpenRouter...<p>$3&#x2F;$15 vs $2&#x2F;$10
        • arcanemachiner6 hours ago
          At what point did Google even offer a decent rug?
          • bpavuk6 hours ago
            the 1.5 through 2.5 generations. you were given quite some generous usage for free in Google AI Studio API. I knew it would break pretty soon, but in the meantime I built a pretty decent reverse-engineering sidekick. LLMs are much better at pattern-matching disassembler output than I am, but someone still needs (yes, <i>still needs</i>) to draw circles and arrows to capture a program&#x27;s behavior.
            • irthomasthomas3 hours ago
              Even early in the 3.1 era the combination of 3.1 pro and Antigravity made an excellent agent. Then they reduced the quota for pro models from very generous down to zero.
    • AtlanticThird55 minutes ago
      This is a great visualization. Thanks for putting it together. If you&#x27;ll placate my OCD brain for a second, it does bother me that Google and Anthropic have the oldest models at the top and OpenAI has the oldest model at the bottom.
    • primaprashant9 hours ago
      As I understand it, Gemini 4 family should be shifted 1 step to the left. Argon &#x2F; Carbon are supposed to be XL and in the same class as Fable &#x2F; Astra.
      • HJain138 hours ago
        They might be weight&#x2F;parameter count wise, but definitely not perf (this is while over indexing on coding though) or price wise.
        • polishTar8 hours ago
          Well... This industry moves fast. Carbon had a higher AA score than Astra and Fable when it was announced, though the other labs have updated their models since. Fable 5.2 is clearly better than Carbon; with Astra 6.2 it&#x27;s a little less clear. I don&#x27;t think there&#x27;s much question it&#x27;s meant to compete against the other labs super-big models.<p>It&#x27;s cheaper than astra&#x2F;fable, but that&#x27;s likely because they know they&#x27;ve been out of the game for awhile and need to earn back mindshare. Also Carbon is pretty token heavy.
          • mnicky6 hours ago
            And another thing is that AFAIK in the past Google replaced lack of skill (relative to Ant&#x2F;OAI) with size (as they had plenty of hw). So Geminis were always the largest model sizes, larger than Opus&#x2F;GPT family. I&#x27;d assume this holds even today, with Carbon vs Fable&#x2F;Astra.<p>Carbon&#x27;s relative slowness would support this as well.
    • satvikpendem9 hours ago
      This is illegible on mobile.
      • wwizo9 hours ago
        Rotating screen 90 degrees seems to fix it
        • vanderZwan8 hours ago
          Not on mine
          • herewulf7 hours ago
            I took the further step of toggling &quot;Desktop site&quot; and then the whole thing displayed beautifully. <i>shrug</i>
          • setsewerd2 hours ago
            I have my rotation locked but still started turning my phone anyway like an idiot. The coffee clearly hasn&#x27;t kicked in yet.
      • smcameron9 hours ago
        Mobile sucks that way.
      • inquirerGeneral8 hours ago
        [dead]
      • scrollop7 hours ago
        Give it to an AI to reformat.<p>It&#x27;s a comment, not a product you paid for.
    • melasadra7 hours ago
      From looking at the model selection in Gemini Web and Antigravity:<p>Flash -&gt; Hydrogen (medium level) -&gt; Argon -&gt; Neon Pro -&gt; Carbon<p>Because Argon and Carbon existed for a while at the same time but treated differently. Pro users get Argon but only Ultra users get Carbon.
    • nosslin8 hours ago
      I like this. I wish the big sites would categorize like this. For most people XL is not needed. M or M+ is good enough for most. If you really are pushing it then go L.
      • bronlund7 hours ago
        I disagree, but maybe I&#x27;m not most people. XL is my daily driver, I can&#x27;t imagine dealing with the mess an M model creates.
        • chrisweekly2 hours ago
          YMMV; IME Opus-5.5 medium is excellent for many purposes.
        • amunozo7 hours ago
          I can only use Fable now as an orchestrator just because it feels so much more intelligent to talk with. Even for silly tasks, it feels so much nicer.
    • tomaskafka7 hours ago
      As someone who stayed away from Gemini models as they were apparently unable to reliably use tools for a long time, they could have named the models with UUIDs and would cause less confusion.
    • KoolKat233 hours ago
      I don&#x27;t know how fair this is to openAI. Astra and Sol cost and speed wise could be in two groups, xl+l and l+m respectively.
    • HJain139 hours ago
      I have based this on model pricing (which I guess also corresponds to their labs&#x27;s idea of performance&#x2F;how they compete) and for google their model cards which specify which model is derived from what
    • xyzsparetimexyz5 hours ago
      Whose idiot idea was it to go for a period table naming scheme? Are we supposed to just think that the heavier elements are the smarter models? That doesn&#x27;t seem to hold up in practice.
    • ganeshsivakumar3 hours ago
      how would you fit in Deepseek
    • EspressoGPT6 hours ago
      Put Astra in L ...
      • HJain136 hours ago
        Why do you say so? OAI clearly pits it against Fable no? Are you talking from actual perf perspective?<p>All of lists like mine are still subjective honestly despite me&#x2F;others trying to make it rooted in as much objective things as possible. For now the current table is what I think reflects that best
        • KoolKat233 hours ago
          Cost and speed. It should be in xl+l
    • ml-anon7 hours ago
      Fable and Astra are not anywhere near the same size. None of the models are priced in a way that is congruent with their size. Honestly wtf is the point in this?
  • impulser_16 hours ago
    Is there any theory on what is Anthropic&#x27;s secret sauce? I have been using the latest Gemini models and they just don&#x27;t seem as good as Opus 5.5 despite the benchmarks. Yeah it comparable on some tasks maybe slightly better but as an overall general intelligent model nothing compared to Opus 5.5 and it hard for me to understand why. Opus feels like human level intelligent like I&#x27;m talking to a coworker basically while GPT and Gemini feel like I&#x27;m talking to an AI and have to change how I talk. It hard to explain and maybe I&#x27;m the only one that feels this.<p>Is it because Anthropic put a ton more effort into the character of the model that it basically roleplays as a human and that why it tends to think better when giving tasks that it might not be trained heavily on.<p>I have used Carbon more than most model to try and get the same results but it just feels like i&#x27;m talking to a smart AI. Which is fine if you are asking questions, researching an idea, but if you actually want to work on something, build something, it just falls apart. Like Opus can run for days on a task and it feels like it understand what it was doing 2 days ago even when it might not be in it context while other models feel like they breakdown and get lost or falter as the task grows.
    • svara15 hours ago
      I think Gemini 4 (Argon is the one I&#x27;ve been using) is very clearly better at writing English than any Anthropic model. It&#x27;s not better otherwise but in that the difference is vast.<p>To the point that I now use a workflow of delegating from Opus or Fable to Gemini4 to write commit messages and PR descriptions, it mostly solves the Claudese problem.
      • qalmakka2 hours ago
        Agree. Claude sucks at writing, you can pinpoint it&#x27;s Claude from a million kilometres away and it always sounds like it&#x27;s writing in slogans or something. Both GPT and Gemini are way better at writing text
      • jjbinx0074 hours ago
        I agree. Gemini writes excellent English and doesn&#x27;t trigger my AI-dar but I can tell Claudish from a mile away. In fact, I just had a notification from Life360 and spotted a Claudism so I now have a strong suspicion they&#x27;re using Claude to assist with code.
      • hikarudo14 hours ago
        I just use Opus with the simple-english skill. No more Claudese.
        • losvedir1 hour ago
          Nah, I haven&#x27;t found complete success with tweaks like that. I would strongly suspect they&#x27;re still there and you might just not be as sensitive to them.<p>I would bet you still get tells like overusing &quot;never&quot; or ubiquitous inanimate subjects with animate verbs (&quot;questions that decide the option&quot;).
        • aloha24363 hours ago
          Simple English skills get rid of the most egregious word choices but their awkward sentence structure still comes through for me. I can definitely still tell, if only because I know what my coworker&#x27;s actual writing sounds like.
          • ignoramous1 hour ago
            &gt; <i>their awkward sentence structure still comes through for me</i><p>My fear is that folks who read Claudish all day everyday will start picking up its style.
        • NikhilVerma9 hours ago
          I&#x27;ve been maintaining SlopSift - <a href="https:&#x2F;&#x2F;slopsift.dev&#x2F;" rel="nofollow">https:&#x2F;&#x2F;slopsift.dev&#x2F;</a> for a while it&#x27;s powered by a dependency parser to deterministically lint English. Then I wrap it around Claude either as a skill or a hook. It&#x27;s really simplified the AI speak for me
          • dgfl7 hours ago
            And yet the prose in the website is still full of claudisms. I’ve been looking for a tool to improve prose, but I still haven’t found one that obviously works. I’m starting to think Claude is just unable to output a sentence that doesn’t sound like itself. Have you run it on your own homepage?
            • spider-mario1 hour ago
              Yeah, I’m curious what it would say about:<p>&gt; Authorable rules inspect the graph for structural tells. Every finding names what matched and the exact text that triggered it.<p>(I don’t know if it’s just me but those are also very much garden-path sentences for me.)
        • Degorath7 hours ago
          Could you link it, if it&#x27;s open? I haven&#x27;t had good success with ASD-STE100 myself :(
          • hikarudo5 hours ago
            <a href="https:&#x2F;&#x2F;github.com&#x2F;AminBlg&#x2F;SimpleEnglish" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;AminBlg&#x2F;SimpleEnglish</a>
        • relistan6 hours ago
          This is better, but you might try the Gemini models if you have access. They are much better writers, even with the simple English skill.
      • relistan6 hours ago
        Yes! All the Gemini models are better writers by a lot. I also delegate PR summaries and titles to Gemini from the other models. Even when you add extra effort to get Claude to humanize, it is still nowhere near as good as Gemini.<p>For planning documents about complex work, Hydrogen does a fantastic job of both the analysis and the writing.
        • cromka4 hours ago
          Do you delegate by hand or thru a skill or some general instruction in your AGENTS.md?
          • relistan2 hours ago
            I use an AGENTS.md instruction
      • disillusioned13 hours ago
        Gemini has reliably been the best writer of the bunch, and Argon has blown me away with how &quot;human&quot; it can sound. It has so much less of the AI smell than the others, though Opus 5.5 definitely lobotomized the absolute sloppery that was Opus 5, at least.
        • antupis11 hours ago
          Yeah usually I use Gemini for writing. Anthropic and OpenAI models even with heavy prompting sound like LLMs where Gemini is way more natural sounding.
        • relistan6 hours ago
          Hydrogen is also quite a good writer, is much faster than Argon, is cheaper, and for most things is very good. Currently one of my favorite models all around.
      • behnamoh15 hours ago
        &gt; To the point that I now use a workflow of delegating from Opus or Fable to Gemini4 to write commit messages and PR descriptions, it mostly solves the Claudese problem.<p>how do you do this w&#x2F;o breaking bank? for Argon to write good commit messages, it must read the diffs and all the work Opus has done, which could easily eat up its usage quota.
        • lilbigdoot14 hours ago
          I use a $20 a month plan with Argon and I honestly find it pretty usable for asking a lot of questions if I don&#x27;t keep it all in one session, or it can do a few small coding tasks a day for me until I hit my limit. I haven&#x27;t hit any limits with the smaller models cuz I don&#x27;t like using a lot of LLM code in my personal projects since me learning a new thing or area by coding is one of the main motivations. Generating messages from diffs&#x2F;notes after the work is done is pretty cheap I imagine.
        • rolosa15 hours ago
          Can you just have opus hand off the changelog to Gemini and have it write from that?
        • relistan6 hours ago
          That would be expensive, but there’s a simpler way. First though: I don’t use Argon for that. The cheaper Gemini models are quite good writers, too.<p>Second: you don’t need to waste a lot of tokens having the other model re-read commits and code. My AGENTS.md&#x2F;CLAUDE.md specifies to write PR summaries and titles and then hand them off to Gemini to clean them up. Very occasionally this requires two trips if Gemini changes something the other model disagrees with, but usually results in very clear and readable PRs, often better than I would have written if I spent 15 minutes on it. And it uses very few tokens.
        • spolitry14 hours ago
          You vastly underestimate size of context the model can process.
      • paganel5 hours ago
        &gt; from Opus or Fable to Gemini4 to write commit messages and PR descriptions<p>Who&#x27;s reading that slop? Humans?
        • cowlevel4 hours ago
          LLMs can read commit messages.
          • paganel2 hours ago
            And pass that information to whom? For what purpose? It&#x27;s slop all the way down?
    • ehsankia16 hours ago
      Every day I&#x27;m just more and more amazed at Opus 5.5 and how ahead it was. Even with the insane pace we&#x27;re going at, it&#x27;s amazing that two months later, we&#x27;re barely putting out models that hit the price&#x2F;speed&#x2F;quality of Opus 5.5<p>No wonder they went for the 5.5 naming, instead of the .1 increase. From just the benchmarks, it seems like neon is basically on par with Opus 5.5 which came out 2 months ago.
      • aenis16 hours ago
        Same here. Spin up a few long horizon tasks, have it manage the whole thing as a project, you can talk to it, interrupt it, engage in side-quests, its like talking to a human.<p>Have another session on another machine (say, IOS build on a mac, and server build on a linux?). No problem, they will happily coordinate with each other through the &#x2F;remote-control protocol, and then you feel like you are in a crowd, with those sessions talking to each other and over you.<p>No such feel with OpenAI nor Google models for me. They are absolutely great for lots of things, but nothing comes close to the feel of Opus 5.5 and Fable.<p>(I am biased; i mainly use the anthropic models so i got used to them).
        • jwpapi16 hours ago
          I dont even feel like fable 5.2 is better. Imho the best model rn is Opus 5.5 MEDIUM.
          • p1necone12 hours ago
            The first time I tried to use fable 5.2 it made a bunch of really amateurish technical design assumptions about my project that cursory reading of docs would have disproven (but also should have just been discarded because of taste reasons anyway). I switched back to Opus 5.5 after that.
        • ravikapoor1018 hours ago
          I had the opposite experience.<p>I had subscription to both anthropic and openai. I was out of credits for openai, so used anthropic to fix some coding issues. Spent whole weekend prompting while it kept failing over and over. Finally on Monday I had openai&#x27;s weekly quota start, 2 hours issue was done.<p>The same thing happened again, after a couple weeks. Two weekends wasted trying to make Anthropic models to behave.
          • aenis5 hours ago
            I&#x27;d think its maybe a question of getting used to a given model, developing good intuitions on how to prompt, divide work and so on? I have spent probably 2000 waking hours this year with Anthropic models, and maybe 200 on openai, so i naturally consider them easy to work with. My own benchmark suite probably reflects the way I work, which is why its mostly Anthropic models getting the top scores etc.
        • ehsankia7 hours ago
          fable 5.1 was strictly worse than opus 5.5 (which was embarasing). Even Fable 5.2 is slightly better, but both slower and more expensive. Again, pound for pound Opus 5.5 from 2 months ago is still probably one of the best models.
          • aenis5 hours ago
            To me the revelation was how good Sonnet is. But yeah, my daily driver for anything is Opus 5.5 and I pray to the gods they won&#x27;t nerf it anytime soon.
      • pjankiewicz16 hours ago
        I agree Opus 5.5 is something different entirely. It is even hard to find tasks for other agents to be honest where you know the results will be picked apart when Opus 5.5 reviews them. In my setup Opus 5.5 uses codex from time to time when I&#x27;m running out of credits and it decided to run 2 astra agents for the same task to improve its performance.
      • nicoburns15 hours ago
        &gt; two months later<p>Mad that this is seen as a long time horizon!
        • ehsankia7 hours ago
          I agree, but we live in an age where there&#x27;s new models announced Weekly. We&#x27;ve had Argon, Carbon, Hydrogen and Neon all <i>after</i> Opus 5.5.
      • Snibz3 hours ago
        Bit of a strange thing to be amazed by. I certainly agree that Opus 5.5 is a strong and useful model, it&#x27;s been the model I&#x27;ve used the most for work in the past few weeks. But is it really amazing? Any other arbitrary AI company could equally come up with arbitrarily advanced models, applications, whatever. I&#x27;m certainly not amazed by the fact that Anthropic provides access to a strong frontier model.
    • nbardy6 hours ago
      I think a lot of it boils down to &quot;constitutional AI&quot;.<p>They did a paper on this a long time ago[1], I&#x27;m sure they have altered their techniques some, but I this was the earlier signal that they really treated the AI as a separate entity and trained it as such. And focus not just on it&#x27;s capability, but it&#x27;s *behavior*, when to push back on the user, etc... Imo. it&#x27;s more about the philosophy of the constitution in this paper than the specific RLAIHF, I&#x27;m sure this idea of &quot;when to follow instructions and when to trust yourself is in their system prompts for all their RL environments, and even their rewards.<p>This idea of an objective function vs what the model should do during human interaction is a huge question and a very subtle one to get right, but it&#x27;s why they were winning in real world application even before they were winning in raw benchmarks.<p>Especially in the era of RL environments dominating more and more compute it&#x27;s vey easy to overfit a model to be very good at accomplishing a narrow task exactly as specified. Much more complicated to have it respond appropriately on turn 5 of a nuanced conversation with a user about their private codebase.<p>[1] <a href="https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2212.08073" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2212.08073</a>
    • foohis16 hours ago
      Anthropic spent a lot more time editing their pre-training data with higher-quality material. For example, they used actual books for pre-training, because books are generally run through human editors for higher quality. (this is from the copyright lawsuits)<p>Other companies would use completely raw internet data for pre-training, which is basically trash. Anthropic&#x27;s secret sauce is that they filtered and edited down their data for only the highest-quality material. You really don&#x27;t need quadrillions of tokens to pre-train, as that&#x27;s just random noise. A smaller-but-higher-quality data set is better in every way.<p>Meanwhile its obvious that companies like Google would use trash data like email and chat logs or Internet forums for pre-training. Just complete trash data.<p>This is a classic case of &quot;less is more&quot; and &quot;garbage-in-equals-garbage-out&quot;
      • hackingonempty15 hours ago
        <p><pre><code> Anthropic spent a lot more time editing their pre-training data with higher-quality material. For example, they used actual books for pre-training, because books are generally run through human editors for higher quality. (this is from the copyright lawsuits) [ ... ] Meanwhile its obvious that companies like Google would use trash data like email and chat logs or Internet forums for pre-training. Just complete trash data. </code></pre> Google has been scanning every book they can get their hands on for 20 years or more. They have a bunch of patents on non-destructive book scanning gear. Surely they are using that data to train their models.
        • foohis14 hours ago
          won&#x27;t help if they also include trash internet text in the training set as well.
          • etoxin13 hours ago
            it&#x27;s kinda known that the gemini models write better english which goes against the &quot;trash data&quot; remarks.
            • Gareth3218 hours ago
              Maybe not. If they’re trained on real world emails and Reddit comments, they’ll write more like “real” people, but they’ll be dumber.
              • rixed8 hours ago
                Exactly my gut feeling as well.<p>It&#x27;s not that Claudese is wrong, it&#x27;s that it&#x27;s so dense as to become enigmatic.
                • perfsakuya5 hours ago
                  Enigmatic language increases the cognitive load on human readers. In this regard, maybe Gemini is better.
            • foohis7 hours ago
              Have you ever tried the Gemini models? They are absolutely atrocious. Like, hilariously bad. One time they kept asking me for &quot;Harriet! Where is Harriet?!&quot; and another time it responded in chinese.<p>They were not trained on good data.<p>And higher-quality source material results in higher-quality emergent behavior, everything from thinking to tool use.<p>Seriously people, read their copyright court case. It literally states why Anthropic focused on books and higher quality material.<p>Google has to learn to not use trash internet content if they want to catch up.<p>(note that I haven&#x27;t tried Neon yet so maybe they learned their lesson and fixed it. All my experience is from 3 months on their Ultra plan and previous models from earlier this year)
        • wahnfrieden14 hours ago
          Surely they are destroying scanned books for training
          • mastax12 hours ago
            They invented those nondestructive scanners before the court cases that effectively mandated destruction.
            • wahnfrieden10 hours ago
              Why does that allow them to use the scans for LLM training? Those scans were used for a different purpose. I don’t understand the logic.
              • yard20107 hours ago
                The logic is: when you have a lot of money, you get a lot of influence and you get the freedom to choose what to do. Be it traffic humans, induce riots, steal property, etc. Oversimplified, but this logic holds most of the time in my experience.
                • crimsoneer5 hours ago
                  No, the logic is that if you have a copy of a book, then create a copy and destroy it, you still have a copy of a book (and so can&#x27;t go and sell on the original).
              • Schlagbohrer6 hours ago
                If I recall correctly its because legally they (Anthropic) can avoid a copyright lawsuit if they destroy a rare book, because digitizing it and then having the original no longer exist in any other form is somehow... it prevents any possibility of a lawsuit due to a quirk in the law.
          • NamlchakKhandro13 hours ago
            [dead]
      • impulser_16 hours ago
        I don&#x27;t think it pre-training, because surely these labs all have similar data and Google should be very good at ingesting, cleaning, and filtering data.<p>I honestly think it just their post-training is fantastic and I think it has something to do with them being very opinionated on how they want the model to act that it basically the best at roleplaying a human. Humans are very good at completely and figuring out things that there aren&#x27;t &quot;trained&quot; on if given basic set of intelligence. So when it working on these long tasks that might not be directly in it training data it still reason through it very well.<p>They got made fun of for this in the pass because they do a lot of research on this including hiring a philosopher.<p>This is just my best guess, because you don&#x27;t see any other labs talk about how they made the model act and Anthropic is basically obsessed with it and talks about it every model release.<p>I think they just have better &quot;taste&quot; and that taste is their secret sauce.
        • tarsinge7 hours ago
          &gt; Google should be very good at<p>Given all the other things Google should be good at and is failing at why would this be the exception?
      • nl15 hours ago
        This is incorrect.<p>Everyone uses books as well as internet data.<p>Anthropics difference is they have a huge data budget for buying RL environments for real world tasks.
        • positr0n12 hours ago
          I get recruiter emails for $150&#x2F;hr to do code review training for anthropic (W-2, although short duration contract).
          • jonah8 hours ago
            For other fields like medical or legal it&#x27;s 2-3x that.
          • nl10 hours ago
            Yes, and if you look on LinkedIn there are many companies doing similar tasks in areas from accounting to physics to HR. They sell to all companies, but Anthropic seems the most aggressive. This from last year:<p>&gt; According to The Information, leaders at Anthropic have discussed spending more than $1 billion on RL environments over the next year.<p><a href="https:&#x2F;&#x2F;techcrunch.com&#x2F;2025&#x2F;09&#x2F;21&#x2F;silicon-valley-bets-big-on-environments-to-train-ai-agents&#x2F;" rel="nofollow">https:&#x2F;&#x2F;techcrunch.com&#x2F;2025&#x2F;09&#x2F;21&#x2F;silicon-valley-bets-big-on...</a>
        • foohis7 hours ago
          They do have a huge RL budget, but their OG primary source were all books they scanned. Books were a lot cheaper and higher quality than other data. This is straight from their court case.
      • mlmonkey14 hours ago
        Microsoft (MSAI) showed that quality &gt;&gt; quantity with their Phi model a couple of years ago. It was a small model, but trained on very high quality data, and did better than much larger models.
      • MattRix43 minutes ago
        I don&#x27;t think this is true because Opus 5 was bad and Opus 5.5 was really good, and presumably those both used the same pre-train? So instead it&#x27;s more of a post-training thing?
      • amarcheschi14 hours ago
        I can&#x27;t say what precisely, but other big companies are also doing what they can to get human data of high quality, just look for all the companies that popped like mushrooms that do ai reinforcement tasks, micro1, ethos, mercor ecc
      • quixoticelixer-16 hours ago
        That doesn&#x27;t really make sense because as good as Opus 5.5 is it writes awful terrible, disgusting slop.
        • weitendorf16 hours ago
          Anthropic doesn&#x27;t want people to use their models for the chat&#x2F;persona style use cases OpenAI has spent more time and effort optimizing for.<p>If you look back at GPT2 and the early days of AI safety&#x2F;risk this was actually something the kinds of AI researchers that joined Anthropic didn&#x27;t want - RL on coding-&gt;using that accelerate development and to bootstrap capabilities across domains was always the plan, not making it easier for people to spam&#x2F;fake homework or trap people in problematic human-AI attention&#x2F;relationship tarpits.<p>I personally find that Opus 5.5 and Fable 5+ are generally fine to chat with but just very wordy and culturally &quot;LessWrong&#x2F;essay-guy&quot;. Which, if a model is going to have a personality, is close to the one that I want (though it does piss me off when it gets preachy about standards Anthropic clearly doesn&#x27;t hold themselves against)<p>I also <i>want</i> to immediately recognize AI generated content on the web or from another person, so I&#x27;m very glad it has a distinct voice
          • quixoticelixer-15 hours ago
            Yeah maybe that&#x27;s the case, and maybe I am more sensitive than other people. But I don&#x27;t really use AI for that sort of thing but even for technical stuff I still find it&#x27;s writing style a real turn off. Gemini still has lots of AIisms but it&#x27;s a lot more tolerable.
            • Degorath6 hours ago
              Have you tried Muse? I only took it for a month on a cheap plan and I did not do much serious work with it, but my initial impressions on the writing style when I wanted it to explain me things was pretty nice.
          • Yizahi5 hours ago
            Discussion about issues with Anthropic LLMs: &quot;Oh, they did that on purpose to thwart some imaginary future problem or legislation, they are genius like that.&quot;<p>Discussion about issues with Google LLMs: &quot;Have you considered that Google are simply bad and failed to do that properly?&quot; (saw this exact line at least a dozen times now on HN)
        • LeoPanthera15 hours ago
          If the training set is smaller, doesn&#x27;t that make sense though? There will be much less variety in its style and so its going to make very recognizable output.
          • quixoticelixer-15 hours ago
            Yeah that could be it. I guess the thing about what makes AI writing so repulsive sometimes is not that it&#x27;s bad like a human would be bad, but that it does the same things over and over.
        • foohis16 hours ago
          It&#x27;s a much, much better writer than literally every other model.
          • quixoticelixer-14 hours ago
            It&#x27;s better than Opus 5 but it still has that incredibly punchy contrasty LLM style that worse than most other models
            • thriving_earthw12 hours ago
              I sometimes use Gemini for quick searches or stuff and the writing style is such a breath of fresh air. I mean I&#x27;ve been so locked in to the anthropic ecosystem because I find that code, the desktop app, and the ability of their models to understand what I mean and execute so well is incredible. But sometimes I&#x27;ll read something it writes and I&#x27;ll want to throw my computer at the wall. I just become apoplectic with rage for a second before I of course cool down and move on.
              • BigGreenJorts11 hours ago
                I&#x27;ve been feeling this more and more lately as I&#x27;ve been <i>trying</i> to delegate technical writing to it. It&#x27;s so horrifically repetitive, verbose and somehow non specific I can&#x27;t even really use it as a base to edit down to something legible.
                • copperx7 hours ago
                  It&#x27;s sad that you can&#x27;t quite delegate the writing task to another model, because Claudese is highly contagious.
      • potamic10 hours ago
        Is this your opinion or you have a source for your claim?
        • foohis7 hours ago
          Not opinion at all. It&#x27;s directly from their court case.<p>&quot;Over time, Anthropic came to value most highly for its data mixes books like the ones Authors had written, and it valued them because of the creative expressions they contained. Claude’s customers wanted Claude to write as accurately and as compellingly as Authors. So, it was best to train the LLMs underlying Claude on works just like the ones Authors had written, with well-curated facts, well-organized analyses, and captivating fictional narratives — above all with “good writing” of the kind “an editor would approve of” (Opp. Exh. 3 at -03433). Anthropic could have trained its LLMs without using such books or any books at all. That would have required spending more on, say, staff writers to create competing exemplars of good writing, engineers to revise bad exemplars into better ones, energy bills to power more rounds of training and fine-tuning, and so on. Having canonical texts to draw upon helped&quot;<p>From: <a href="https:&#x2F;&#x2F;copyrightalliance.org&#x2F;wp-content&#x2F;uploads&#x2F;2025&#x2F;06&#x2F;Bartz-v.-Anthropic-Order.pdf" rel="nofollow">https:&#x2F;&#x2F;copyrightalliance.org&#x2F;wp-content&#x2F;uploads&#x2F;2025&#x2F;06&#x2F;Bar...</a>
      • nikhilsimha9 hours ago
        i saw a researcher at uw claim that claude is trained on genome sequences and that others arent.<p>genome sequences are probably very dense in “reasoning”
        • oasisbob9 hours ago
          I have a weak theory that some of the strange contortions in Claude come from training on scientific literature.
          • __s2 hours ago
            Was recently trying to read &quot;categorization is baked into the brain&quot; as someone who doesn&#x27;t read papers it tickled my brain similar to the hours of slop I read per day
      • olalonde13 hours ago
        Where do you get that information from? Or is it just speculation?
      • UltraSane15 hours ago
        I think by now everyone training trillion parameter models is using almost every book, magazine, research paper, etc they can get.
      • mrheosuper12 hours ago
        Is that reason why Claude talks so like .... a nerd.
      • Topfi15 hours ago
        Purely outside looking in, but yes, I&#x27;d argue you are right and higher-quality is a major part of it, both for pre- and post-training. There appears to be a far greater willingness to adopt specifically tailored training sets from third-parties, consistently update and build on them.<p>From simple frontend work to strategically adding virtual rulers when working with LIDAR scans to extract more accurate data, there are very clear points when Anthropic introduces a new external data set into their mix and while I have seen the same from competitors such as Google, there simply appears to be a greater tendency to stick with these once they have them.<p>As an example, at release it was quite apparent that Gemini 3 Pro, 3 Flash and Opus 4.5 had a similar, possible even the same frontend-focused training base bought in from a third party. But while Anthropic has had both smaller and larger improvements across each subsequent Opus release, I have not noticed such a consistent improvement with Gemini models over the last year. Maybe Anthropic simply buys updates more often, maybe they create more artificial PRs on top, who knows.<p>Same with the LIDAR scans, there was a point a few months ago when models started to exhibit a very systematic way of handling visual references in some few leather-crafting tasks, enabling whole new ways of working more efficiently combining existing measurements with images. Step-by-step pixel measurements, compensation for angles and poor captures, etc. Have since seen this in other models, again very much feels like a bought in data set that they are pushing as far as it goes.<p>Then there is 3D capability with models like Muse Wonder Full. I have not used it myself, if it ain&#x27;t on API, it doesn&#x27;t exist, but have seen PRs and do find the model output very parsable, despite it being terrible at any modelling task because these capabilities do not emerge from just training on vast amounts of data, they require high-quality, well tailored, step-by-step reference data of the entire process.<p>Harness wise, while pure output performance is very consistent across harnesses in my evals (I nowadays use both OpenCode via API and the labs own harness three times each to normalise), Anthropic also appears to be far more willing to use the time between releases to prepare features for the next. Stuff like workflows, proper subagent implementation, advisor, etc. often have found there way into Claude Code either in-between models or were first seen with a model that couldn&#x27;t fully leverage these. Also feel that explained the good overall launch experience Fable 5 delivered, they simply used the time between announcing Glasswing and launching full access to get the UX&#x2F;DX polished.<p>Meanwhile, I am still testing GPT-6.2 Astra and Sol so will refrain from making any more judgement on either till I have all data, but apparently there has been an extensive last minute rewriting of the system prompt to improve performance and &quot;behaviour&quot;. Given the gap between GPT-6 Astra and yesterday, along with statements by OpenAI staff alluding that they had this model ready for a while, such things should happen in a far more structured, systematic, well thought out manner, not last second after some negative outside feedback. Heck, from what I know Anthropic, unlike any other lab, does zero early access outside the limited hardening for very exclusive partners like Glasswing. Certainly no early access for regular users, reviewers or the like from what I&#x27;ve heard, so Anthropic does all that prep-work for model releases with less feedback than other labs.
    • charm1376 hours ago
      Their models simply overthink - they are simply some of the most token inefficient models from the frontier labs: <a href="https:&#x2F;&#x2F;artificialanalysis.ai&#x2F;?intelligence-efficiency=output-tokens-per-task&amp;models=claude-sonnet-5-5-medium%2Cclaude-sonnet-5-5-high%2Cgpt-6-2-sol-xhigh%2Cclaude-opus-5-5%2Cclaude-fable-5-2%2Ckimi-k3-1-pro%2Cgpt-6-luna%2Cgpt-6-2-astra%2Cgemini-4-neon%2Cclaude-haiku-5-5%2Cgpt-6-2-sol-medium%2Cgemini-4-carbon%2Cclaude-sonnet-5-5-xhigh%2Cgpt-6-2-sol%2Cgpt-6-1-terra%2Cclaude-sonnet-5-5%2Cclaude-sonnet-5-5-low%2Cgpt-6-2-sol-high%2Cgemini-4-hydrogen%2Cmuse-ember%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5-medium%2Cdeepseek-v4-1-flash-1017%2Cclaude-opus-5-5-high#intelligence-efficiency-tabs" rel="nofollow">https:&#x2F;&#x2F;artificialanalysis.ai&#x2F;?intelligence-efficiency=outpu...</a>
      • mrklol6 hours ago
        If that’s the current way to get the best results, I wouldn’t see that as a problem
        • charm13746 minutes ago
          It isn&#x27;t the best way, as is visible from the Pareto frontier curves here for GPT.<p>Secondly, this strategy leads to Claude models also being the slowest models - so each response &#x2F; tweak iteration wastes more time than other models.<p>Thirdly, because Claude&#x27;s intelligence drops off so rapidly with thinking levels, you end up paying way too much for marginally higher intelligence (which may not even be perceptible) for its Max levels whereas the equivalent levels of its competitors cost less due to fewer output tokens being used (and generally saner pricing).
        • paultnylund5 hours ago
          [dead]
      • UFxESTER5 hours ago
        [dead]
    • cavisne10 hours ago
      My theory is firstly its focus - they got to skip to GPT 4 (ie clean room implementation by the same people), they never cared about consumer or video gen or image gen.<p>Secondly, they somehow made all this work on public clouds with really crappy (compared to baremetal nvidia) networking. I think thats probably a lot of post training (which doesn&#x27;t need such a large&#x2F;tightly connected cluster), and some architecture that is less reliant on a tightly connected cluster for pretraining. This would also explain why they have been able to use Trainium, and somehow beat Google while using TPU&#x27;s.
    • surround16 hours ago
      I suspect that Anthropic has more user data to train on from Claude Code. Claude Code became popular sooner than Codex and Antigravity, and remains more popular - thus more data to collect.<p>Cursor was popular before any of those three, and just look at how much Grok improved after they acquired Cursor&#x27;s data.
      • disgruntledphd28 minutes ago
        &gt; Grok improved after they acquired Cursor&#x27;s data.<p>I never used Grok before the Cursor acquisition (we use Cursor mostly), and if this is improved, I shudder to think what it must have been like before the acquisition.<p>It&#x27;s like an irritating Sonnet 3.5.
      • behnamoh15 hours ago
        &gt; Cursor was popular before any of those three, and just look at how much Grok improved after they acquired Cursor&#x27;s data.<p>No one that I know of uses Cursor anymore, and Grok has been disappointing since ages.
        • teknologist5 hours ago
          Grok Bot seems relatively popular, and it does launch and use Cursor and Cursor Agents.
        • surround15 hours ago
          I don&#x27;t disagree.
    • ozgung2 hours ago
      &gt; Is there any theory on what is Anthropic&#x27;s secret sauce?<p>Yes, it&#x27;s called the &quot;Soul Document&quot;. A long trade secret document depicting everything about Claude (behavior, values, character, boundaries etc.) in great detail. They use that in training to shape Claude&#x27;s identity.<p>I think this makes all the difference because I feel like talking a very good coworker. It understands me, helps me and slightly pushes me to be a better version of myself. This is in contrast with a robot blindly doing what I want.
    • vintermann6 hours ago
      There are a lot of real world specialized tasks you can&#x27;t get better at by logical reasoning or role playing as a human. How you conjugate in some small southeast Asian language. What handwriting looked like in 18th century Norway. These are the tasks which first degrade when you shrink a model. Google&#x27;s models, in my experience, do well on them, other models not so much.<p>It makes sense too, because of course Google should have a ton of exotic training data. It&#x27;s probably their comparative advantage, over getting super-clever with reinforcement learning.
    • socalgal28 hours ago
      I&#x27;m personally having amazing luck with Gemini Argon at work (what our company chose). I use Claude at home and have had great luck there too. But, they are much more complicated issues at work than at home so I don&#x27;t really know how to compare them. I tried a few simple graph one-shot demos and Claude (Opus 5.5 - medium) did better than Gemini Argon medium for those small things. But for big things on our company&#x27;s C++ code base, Gemini Argon has been amazing. I don&#x27;t have Claude there to compare.
      • DetroitThrow8 hours ago
        Argon seems really good at C++ in particular in my experience. I prefer Claude for Rust, though.<p>They both seem to break down at similar size codebases or complexity levels, with maybe Opus having a slight edge over Argon.
    • thatxliner12 hours ago
      &gt; Is there any theory on what is Anthropic&#x27;s secret sauce?<p>I&#x27;m planning to write a blog post on it, but my theory is that it&#x27;s aligned to not be &quot;politically correct&quot; but instead to have a solid stance. In the Claude Constitution, they explicitly say how they train it to have judgement and not &quot;follow rules&quot;
    • thanhhaimai16 hours ago
      It&#x27;s likely that the agent configs you have might overfit to Claude models.<p>I&#x27;d recommend trying Gemini Neon on Antigravity, with a trimmed down (or new) agent configs.
    • m_ke15 hours ago
      Anthropic has a proper feedback loop with claude code. They see way more real production data and can detect where their models are weak.<p>So they can train on prod traces or at the very least use user prompts to seed good synthetic tasks to train on.
      • chilmers15 hours ago
        OpenAI has plenty of feedback from Codex. Maybe not quite on the same volume as Claude Codex, but still massive. So I&#x27;m not sure that&#x27;s the whole story.<p>I doubt there&#x27;s a single reason for why Anthropic are ahead right now. Building a frontier model requires executing well in a wide variety of things. It&#x27;s probably a result of being slightly ahead in various aspects of of data collection, model architecture, pre and post training, harness development, etc.<p>It&#x27;s a shame that so much of what&#x27;s happening is being closed doors. Maybe one day, decades from now, somebody (or some AI) will write a definitive history of this period, explaining what was going on at the various labs, how their research compared, and what the major players really thought.
        • cmrx642 hours ago
          the businesspeople more prone to using codex are liable to avoid contribute their data to training. whereas claude has endless amounts of enthusiasts.
    • antupis11 hours ago
      It depends for programming Opus&#x2F;Fable is the best but eg for code reviews or documentation Carbon is better. I think here Google own code review&#x2F;documentation practices and thus probably best dataset in whole world really pays dividends.
    • porphyra13 hours ago
      They just have an incredible data flywheel I think. Since they were first to market with claude code, they have more training data than others. They also pay a lot ($150 per hour) for SWE experts to train the AI. I get some random recruiting emails occasionally from companies affiliated with them.
      • mattlondon6 hours ago
        I think all the big labs do this - the Scale AI company that Meta paid squillions for was basically a company that recruits experts for human training for the big labs.
      • jbonatakis13 hours ago
        Yeah I’ve gotten the same thing. If anyone from Anthropic is reading this: it’s not enough to make me quit my job for a short term contract. Try doubling it and we’ll talk
    • mattlondon6 hours ago
      For what it is worth, I find Opus really, really irritating to use. Just the way it talks is really annoying.<p>If they&#x27;ve done anything with their special sauce, it appears to be to train it on the behaviours of jerk engineers who are smarmy, overconfident, and arrogant.<p>I prefer to use the Gemini models - if I need to spend 8 hours a day talking to something, I dont want that something to be sanctimonious Opus models.
    • irthomasthomas14 hours ago
      This model matches Fable 5.2 for about 40% the cost. Not sure how you can take from this that anthropic has some secret sauce?
      • cloverich14 hours ago
        According to benchmarks; time will tell. I see a lot of comments here and on X about how Google&#x27;s caught up. But when Opus 5.5 came out, I didn&#x27;t see comments. I saw examples of what people built, and it was mind blowing. Stopped the world for a few days for me and others, while we re-configured our brains around what was possible. I&#x27;m entirely willing to believe Google&#x27;s catching&#x2F;caught up but - lets have some real people build and show off like they did with Opus and Astra.
        • DashAnimal11 hours ago
          I&#x27;m seeing a lot of cool stuff on twitter already.<p>For instance:<p><a href="https:&#x2F;&#x2F;x.com&#x2F;SPAC89&#x2F;status&#x2F;2123171830739632189?a=WIEEh4US46MtegvY7Z4&amp;s=20" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;SPAC89&#x2F;status&#x2F;2123171830739632189?a=WIEEh4US46...</a><p><a href="https:&#x2F;&#x2F;x.com&#x2F;chetaslua&#x2F;status&#x2F;2122705177203642485?a=7aENHW-m2xfnhsIkfAM&amp;s=20" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;chetaslua&#x2F;status&#x2F;2122705177203642485?a=7aENHW-...</a> (woops this was carbon, but still cool so I&#x27;ll keep link)<p><a href="https:&#x2F;&#x2F;x.com&#x2F;kimmonismus&#x2F;status&#x2F;2123168299890704735?a=IWO2O6pK4NTAzM1aXe4&amp;s=20" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;kimmonismus&#x2F;status&#x2F;2123168299890704735?a=IWO2O...</a>
    • jorl1716 hours ago
      Here&#x27;s my average experience with Carbon.<p>(After a long conversation with existing messages)<p>User Prompt: Ok, so SSH to the unraid server and update the plex db<p>Reply:<p>I need to think how to answer this.<p>The user is asking me to SSH into the unraid server. This seems to imply there is a machine to which we can SSH to. I should look for places where hosts may exist.<p>The implication is that once SSH is established to unraid, the plex db can be updated. It seems plex is a service that the machine has in which we can run updates.<p>I&#x27;m wondering if plex might be the Plex service the user has been discussing.<p>So the user doesn&#x27;t want to SSH to the server - actually they want to SSH to the server.<p>... Formulating a plan ...<p>[some messages later]<p>Aha! THIS IS THE SMOKING GUN! `~&#x2F;.ssh&#x2F;config` HAS THE REQUESTED HOST!!! IT&#x27;S TIME TO CONNECT!!<p>[some messages later]<p>Ok, time to act. I&#x27;ll ssh to the server and follow coherent rules.<p>----<p>I am shocked at how primitive its reasoning sounds (it reads just like gemini 2.5-pro&#x27;s old reasoning). It often gets the job done (although it&#x27;s quite slow and can&#x27;t tackle more sophisticated problems).<p>Perhaps Anthropic&#x27;s internal reasoning is also like this, but they really rewrite it to make it seem more coherent (which certainly sells it on me)? I almost want to click stop just by reading this nonsense.
      • iamdelirium16 hours ago
        All displayed reasoning is not real reasoning for all models now. It&#x27;s fake since they can be trained on.
        • jorl1716 hours ago
          I want to believe you, and it&#x27;s what makes the most sense! But if that&#x27;s the case, why is google (intentionally?) presenting such terrible reasoning traces? To make sure that if they&#x27;re trained on, they&#x27;re trained on garbage? I guess it makes sense, but still...<p>I was not really fabricating the example I gave too much. Here&#x27;s a verbatim transcript from a session which was filled with this stuff:<p><pre><code> ... thought LET&#x27;S READ RecordReader.cs RIGHT NOW!! RecordReader.cs IS WHERE EVERY SINGLE TABLE, ReadPlan, DecideValueVisibility, ReadRowCounts , AND ShowValues READS ITS DATA!! Let&#x27;s read RecordReader.cs (lines 1-300)!</code></pre>
          • 8note16 hours ago
            reasoning traces also arent real reaaoning. its what tokens warm up the outpt the best
          • dist-epoch16 hours ago
            What you&#x27;re seeing is a small model summarizing and possibly obfuscating the real thinking.
            • jorl1715 hours ago
              Yes, I understand it.<p>It seems rather ironic that Anthropic givens the most logical reasoning traces, though. You&#x27;d think if they&#x27;re ahead they&#x27;d either hide them or make them produce garbage (like what seems to come out of Google&#x27;s models).
              • orbital-decay15 hours ago
                Neither has any relation to the actual reasoning trace. The summary is actively misleading and you&#x27;d be surprised how different it is from the raw CoT of the model. Just ignore it.<p>If you want to see example traces without jailbreaking claude or gpt, see open models that display the CoT without trying to hide it. It&#x27;s pretty useful sometimes, with usual caveats (it deceptively looks like English but is heavily reward hacked, and it&#x27;s generally a weak proxy for model&#x27;s reasoning process).
                • jorl1715 hours ago
                  This makes sense. Nonetheless, I find that reading Anthropic&#x27;s reasoning traces is sometimes a trigger for researching more stuff. Often I&#x27;ll see models thinking (or fake thinking, whatever) about something and I&#x27;ll be curious to check out more about that topic if I&#x27;m not particularly familiar with it.
                  • jrowen8 hours ago
                    FWIW, you&#x27;re not wrong. As I said in my other comment, I don&#x27;t really know what other models are up to but Claude is definitely demonstrating a deep level of understanding, and I learn things from it all the time.<p>Sometimes I&#x27;m like...is everybody else using the same thing that I am? This is insane.<p>They definitely let it churn until it figures something out, it&#x27;s not uncommon for it to take 15-20 minutes to answer, is that normal? They just have &quot;relentlessness&quot; maxed out and some kind of second-order structure to orchestrate a bunch of agents into one unit of work.
                    • jorl178 hours ago
                      That&#x27;s the exact same feeling I have.<p>It&#x27;s obvious that what we&#x27;re getting are summaries. Sometimes I think I even see vestiges of the summarizing model instructions. But, that being the case, it&#x27;s bizarre why Gemini&#x27;s summarizer acts like a toddler, and Claude&#x27;s actually lets me learn and, more importantly, stop it at its tracks when I see it&#x27;s going down the wrong path.<p>On the other end of the spectrum, if you toy around with the open-weights models, you see the insanity that is going on in their output. I can totally believe in a scenario where claude is basically doing that and then summarizing it into something coherent, getting rid of the 10 opposing viewpoints, &quot;aha!&quot; and &quot;Hmmm, but wait!&quot;, but it does it elegantly.<p>I&#x27;ve tried constantly to use other models (and I do), but Claude spoils me to no end. I keep thinking I must be holding everything wrong.
                      • disgruntledphd25 minutes ago
                        &gt; On the other end of the spectrum, if you toy around with the open-weights models, you see the insanity that is going on in their output. I can totally believe in a scenario where claude is basically doing that and then summarizing it into something coherent, getting rid of the 10 opposing viewpoints, &quot;aha!&quot; and &quot;Hmmm, but wait!&quot;, but it does it elegantly.<p>Yeah, the open weight models (the only ones that show the reasoning tokens) are just absurd. Wait but why oh wait now I have it, this makes total sense!<p>I find it kindof endearing, tbh as I definitely have similar trains of thought when developing.
                • deaux14 hours ago
                  Off-topic, but this kind of conversation tells me just how little the average HNer still uses LLMs outside of specific coding tasks, or at least pays attention to them. Which in turn explains so many comments. Probably a good thing though.<p>You&#x27;re of course completely right. Interestingly the &quot;summaries&quot; have only become more diverged over time. I now believe the big 3 are barely summarizing at all any more. I deem it likely that the small models don&#x27;t get the CoT at all, they just completely make up something &quot;plausible&quot; solely based on the user message with a little bit of previous context. Much of the time it will be the complete opposite of something I asked for, or barely related.<p>Anthropic has changed this up multiple times over the last 12m, I&#x27;m talking about the iteration that was around the longest this year. I think they changed it again within the last feww weeks and potentially started actually summarizing again.<p>Your suggestion is the right one. Everyone who uses LLMs for their job should spend some time with open models like Deepseek and see what the thinking actually looks like.
                  • marmarama6 hours ago
                    &gt; Everyone who uses LLMs for their job should spend some time with open models like Deepseek and see what the thinking actually looks like.<p>I&#x27;m not sure DeepSeek 4.x is a good example; its reasoning traces are extremely verbose and detailed. Which is probably one of the reasons it performs so well, and appears to be a deliberate architectural decision to improve skill at the cost of token usage and time-to-answer.<p>A better comparison might be Gemma or Qwen. Running these locally and with no summarisation, the reasoning traces are not all that different to the big proprietary models in terms of verbosity and content, even if they&#x27;re not as smart.
                • jrowen8 hours ago
                  I mean, it does though. The sequence of thoughts that Claude goes through in response to my queries blew me away the first time I used it, and it continues to impress. Just the act of putting those sentences together demonstrates such a deep understanding of what I&#x27;m saying. If it&#x27;s just an approximation, whatever it&#x27;s actually doing is even better.<p>I haven&#x27;t even really tried to use other models in the same way, and this thread makes me feel like I don&#x27;t need to.
        • behnamoh15 hours ago
          Not only that, Anthropic injects fakes reasoning and tool calls in there to make it harder for distillers. It&#x27;s shitty behavior but you&#x27;d do the same if you had an edge you wanted to maintain.
      • aenis16 hours ago
        One of my long horizon tasks benchmarks involves writing logic controlling a team of soldiers in a simulated environment. Argon and Carbon both made such discoveries such as &quot;it would seem that i need to be pointing at an enemy to shoot them&quot;. Its almost entertaining to read, its like its discovering the world on first principles.<p>(And it did produce a very competent algorithm, though it took a few days and a metric ton of tokens; I just triggered the same task with Neon).
        • jorl1716 hours ago
          I&#x27;m curious about how it goes with Neon! This night I&#x27;ll leave Neon with some lofty goals on a project and I&#x27;ll see how it fares too.
        • aatd8614 hours ago
          lol Aenis the god.
    • jaccola16 hours ago
      Claude annoys me with condescension sometimes... makes it feel far less like a &#x27;tool&#x27; and can even come across as a bit judgemental!<p>Just today I was filling out a security questionnaire and it said &quot;I won’t write a PDF that claims zero credentials while this token is in the logs.&quot; But we were in the process of removing this from the logs so it was perfectly accurate to write and I wish it had deferred to my explicit request...<p>I think Anthropic overcorrected on the sycophancy; Google models to me seem a nice balance of highlighting issues &#x2F; risks while still doing what I actually ask.
      • aenis16 hours ago
        Anthropic models won&#x27;t lie. When we fix security issues identified with sth. like Wiz and are writing a report, it won&#x27;t write a summary suggesting that something we just fixed wasn&#x27;t broken (even if that thing never made it to production). It has something akin to moral standards :-)
        • willis93616 hours ago
          I&#x27;ve seen them ignore clear directions that in documentation it&#x27;s supposed to read after a compaction. That seems a lot like lying to me.
      • enraged_camel16 hours ago
        Claude models will go to <i>extraordinary</i> lengths to ensure sensitive data does not end up in their context window. It&#x27;s a bit nuts. But frankly I appreciate it a lot. Even when it is annoying I know the model is doing the right thing.
        • hadlock15 hours ago
          Opus 4.x really struggled with secrets rotation in sops, even in 5.5 it&#x27;s constantly emitting secrets to the shell history. Thankfully everything gets tested in dev, but Claude models are the reason why we have such good secret rotation now.
        • impulser_15 hours ago
          Yeah, I notice this as well. I was building something to help me understand logs from the different coding agents and it wrote redaction code to redact secrets out of the logs just in case if the logs had any so it wouldn&#x27;t see them.
      • mattlondon6 hours ago
        [dead]
    • tekawade7 hours ago
      I mostly use AI with coding. IMP Claude Code is the secret sauce. It had pioneered and now continues to evolve many new concepts. Which are later adopted by others.<p>Maybe someone used all with API and goose or open code can chime in here.
    • epolanski16 hours ago
      &gt; Is there any theory on what is Anthropic&#x27;s secret sauce?<p>Lots of PR it seems?<p>I have all of Ant&#x2F;OAI&#x2F;OpencodeGo subs and test lots of models and I just cannot understand how and why some people find specific models so ahead, when it seems to me like reality is so nuanced and so specific to the particular task and how you approach it.
      • novaRom4 hours ago
        They have tripled their PR team during just 2025 to roughly 80 people, today numbers can even be larger.<p>Btw, just one B200 x 8 server is all any mid-size company needs: any open source harness + any 80 - 120B active params model - it will fly and serve hundreds of agents. All data and processes are yours - no subscriptions, full control.
      • deaux14 hours ago
        More like, most people especially on here do 90% of their &quot;tests&quot; on very similar tasks - coding on specific codebases - and the other 10% is just asking stuff in chat. So they never get to experience the jagged edge. Anthropic models could be horrific at translation to Nahuatl or greenhouse cherry tomato management compared to Gemini and OpenAI but there&#x27;s one person on HN doing such things and tens of thousands doing enterprise SaaS coding.
      • WarmWash16 hours ago
        Never underestimate the power of minimalist hand drawn graphics on an off white background with a clean typeface.
    • BetterThanSober7 hours ago
      &gt; Anthropic&#x27;s secret sauce<p>For casual use it&#x27;s the harness for me, using Gemini&#x27;s web interface feels like I&#x27;m fighting with all the options while on Claude&#x27;s I just shoot and let the AI activates whatever tooling they need
    • danduma13 hours ago
      You are not alone. I have the exact same experience: opus just <i>gets it</i> and the others feel like cheap toys
    • neximo6413 hours ago
      Exclusivity on RL environments they license.
    • baq16 hours ago
      Opus 5.5 leads the CanTellForestFromTheTreesBench 42.0.
    • e12e14 hours ago
      How do you evaluate opus 5.5 as &quot;better&quot;? My experience has been that Argon was ok - maybe equivalent to opus - but carbon in orchestration mode was finally able to rewrite some early experimental (ai slop) front end code - and break a bit out of UX&#x2F;UI rut that a couple of projects were stuck in.<p>It might be that opus is better <i>in</i> better structured projects, and carbon is better at (re)structuring poorly structured projects?<p>Carbon is a better project manager&#x2F;product owner - opus a better programmer?
    • atemerev13 hours ago
      The secret sauce is sandwich statements. Just look at any Claude response. Positive-negative-positive. Bamm, and we are hooked.
    • scotty797 hours ago
      &gt; Opus feels like human level intelligent like I&#x27;m talking to a coworker basically while GPT and Gemini feel like I&#x27;m talking to an AI and have to change how I talk.<p>It&#x27;s possible that after 4o OpenAI has RL to make its models less human.
    • sudohalt12 hours ago
      There isn&#x27;t because Opus 5.5 isn&#x27;t better. There is no objective reason to believe Opus 5.5 is better.
    • risyachka14 hours ago
      Results usually reflect company culture and with big companies no matter what you do - if the culture is not there - you are doomed.<p>Google is the best example of this with their product graveyard. No matter what they release- they just can’t make a good&#x2F;nice&#x2F;usable product.<p>Just check google ai studio as one of latest examples.
      • spolitry14 hours ago
        AI studio isn&#x27;t great, but what is a better lightweight studio product than AI studio?
  • declanjackson16 hours ago
    Something that is interesting is how it&#x27;s positioned in the overall fleet of Gemini models. Although overall more Intelligent than Carbon, Carbon is still positioned as the &#x27;long horizon&#x27; model. This is clear on AA Index, where we see higher overall scores for Neon but lower scores in evals like CritPt which test research-scale reasoning capabilities.<p>With the frontier getting more jagged, we might see more of this &#x27;less intelligent overall but useful for the frontier tasks&#x27;-type models, meaning picking a model will only get more difficult from here.<p>(FD: I work at AA)
    • GodelNumbering16 hours ago
      &gt; (FD: I work at AA)<p>I do apologize in advance for putting you in spot like this, but there are two things I have been quite curious about.<p>1. Since 60% of the benchmarks contributing to AA are now private only, how do you guys ensure transparency? More precisely, how can someone verify the transparency in scoring?<p>2. There are also disclosures about custom benchmarking services offered to AI labs, what does AA do to keep the custom benchmarking services not getting access to the same&#x2F;similar data that is used in the private benchmarks?
    • seizethecheese16 hours ago
      Okay, this &quot;better at long horizon tasks&quot; thing might explain why Carbon scores worse on most benchmarks yet is the larger model.<p>I looked into this using data I have from a battle system built on top of OpenCode (so we&#x27;re trying to capture real world use). It looks like Carbon actually does a lot better when it uses significantly more output than opponents.<p>Win rate when output is less than 2.5x opponents: 53% (41&#x2F;78)<p>Win rate when output is more than 2.5x opponents: 73% (41&#x2F;56)<p>The p-val on this is 0.02. Data from <a href="https:&#x2F;&#x2F;agentduel.com&#x2F;leaderboard" rel="nofollow">https:&#x2F;&#x2F;agentduel.com&#x2F;leaderboard</a>.<p>There doesn&#x27;t seem to be any relationship between actual size of task though. Win rate appears random based on total output of Carbon and also opponents. (Though our data is still small, so it might exist.)<p>Looks basically like Carbon was trained to think a lot longer, not necessarily that it&#x27;s better on large tasks.
    • abixb16 hours ago
      Yes, I feel the next differentiation in model capabilities would be wrt how &#x27;long of a horizon&#x27; goal they could autonomously work toward without breaking. We&#x27;re in need for a new generation of evaluations and benchmarks for this -- METR doesn&#x27;t cut it anymore.
    • seizethecheese16 hours ago
      That would clear up why Carbon is such an outlier in terms of output tokens per task.
    • WarmWash16 hours ago
      Neon is probably distilled from Google&#x27;s internal successor to Carbon.
    • drivebyhooting16 hours ago
      What we need is a model that automatically delegates intelligently (via RL and benchmaxxing?) so I can be freed of this burden.
      • baoooooooooooo16 hours ago
        This comment always comes up but isn’t this literally irreducibly complex in the sense of Kolmogorov complexity?<p>I’m out of my depth here but it does seem prompts have a “smell” to their complexity that means we can use a classifier to mate them to a model with better success than random noise, but still don’t essentially all of the savings associated with systems like this just come because there is some “complexity arbitrage” where there is enough of a token cost differential between models that you always save money by trying the cheap model first and then trying a hard model if it turns out to be hard?
        • Gareth3218 hours ago
          I’m also not an expert but that sounds a lot like the toilet seat problem. Mathematically it’s preferable to just always leave the seat as you used it, instead of using effort to return it to some desired state. This minimises effort in aggregate for all parties. I think it’s possible that routing every request through a simpler model first ends up more expensive in the long run, depending on the cost of that simple model. And if it’s too simple, the routing inefficiencies and mistakes will produce worse results.<p>I think one of the more difficult problems with this is that the complexity is often unclear at first. It’s not until some thinking and analysis that the complexity emerges. It’s just not possible for a simple model to deduce this, and we don’t want a dumb model investigating a complex issue.<p>I think in the medium term we end up with large models which are efficient with simple queries. Meaning very few tokens used to answer simple queries, keeping costs low. So one just uses this large model for everything.<p>Longer term I think LLMs move closer to the structure of the brain, whereby different regions&#x2F;models have specialised functionality, and they communicate and coordinate depending on the task. For example, telemetry doesn’t need to pass through the prefrontal cortex. It just goes straight to the telemetry model which is super simple and cheap.
          • ethbr11 hour ago
            &gt; <i>Mathematically it’s preferable to just always leave the seat as you used it</i><p>Ah, but the consequences of a female type partner encountering an up seat in the middle of the night far outweigh my consequences of having to raise the seat during the day.
            • Gareth32150 minutes ago
              This is a good point. If your partner is blind and disabled the extra effort may be warranted.
      • iamdelirium16 hours ago
        Antigravity has a Orchestrator model that does this.
      • nicoburns15 hours ago
        This is being done at the layer above. &quot;routing&quot; is the term to google.
        • drivebyhooting15 hours ago
          Routing at the prompt level seems naive. How is that better than a janky classifier.<p>What I want is a deeply introspective model that knows to take a step back and call in for aid.
          • ethbr11 hour ago
            &gt; <i>a deeply introspective model that knows to take a step back and call in for aid</i><p>This would mean training&#x2F;RL a model to know it&#x27;s getting nowhere and call for help.<p>But afaik, the only way a model would be able to do that would be review of its own historical output.<p>And if that&#x27;s what&#x27;s being reviewed anyway, doesn&#x27;t it make more sense to just layer a separate specialized &quot;anti-loop&#x2F;stuck&quot; model on top to judge the output trends?
          • nicoburns14 hours ago
            It&#x27;s being done at the harness level. It&#x27;s still a model making the call.
      • refulgentis16 hours ago
        There&#x27;s stuff that does this and I&#x27;ve put a substantial amount of time last couple months looking into it, unfortunately you always, always, end up at chicken-and-egg problems.<p>It&#x27;s really frustrating to describe. Easiest is you can&#x27;t score on just the prompt, you need the full context. Ex. &quot;Solve Fermat&#x27;s last theorem&quot; initial prompt then &quot;continue&quot; followup prompt. Which gets you back in the tarpit of &quot;if I have to pay a model to do an inference run on context, then, the context is cached and I should continue using that model.&quot;<p>The vision of another model thats RL maxx&#x27;d on knowing how to delegate has a few problems. One, we can observe externally, it died&#x2F;didn&#x27;t work even at OpenAI&#x27;s scale and willingness to make it work, i.e. the...GPT-5.0? IIRC...router that silently went away. The other, if such a model was built, it&#x27;d also be good at regulating <i>it&#x27;s own</i> thinking enough that it&#x27;d naturally use less tokens.<p>In practice, the frontier hackers have settled on a decision model (i.e. Jev) x calibrating effort level or choosing from a set of models x reasoning effort tuples they specify rules for choosing in their Jev prompt. My guess is this stays stable and the only real choice.
        • Gareth3218 hours ago
          I suspect you are correct but damn if that’s not a deeply unsatisfying conclusion.
      • canadiantim11 hours ago
        Aren&#x27;t people using jev or similar for decision-making like this?
      • ihsw16 hours ago
        [dead]
    • chr15m11 hours ago
      Does this mean models are not generalizing like the labs hoped?
    • mi_lk16 hours ago
      what&#x27;s AA
    • s__j7 hours ago
      [flagged]
  • bayindirh4 hours ago
    Many people rightfully compare these models on code generation and other long term tasks. Very few people are using these for researching things and analyzing what they have found so far from the sources, per my observation (I might be wrong, though).<p>On that latter category, I find Gemini superior to Claude for many tasks. Of course Google has an advantage when it comes to reaching places in the internet via its index and things like YouTube, but Gemini is really more useful than people give credit for the tasks I use it for.
    • melasadra4 hours ago
      Agreed. I enjoyed Argon and Carbon for general reasoning especially when it&#x27;s used for context outside the US e.g. in Southeast Asia.<p>Claude&#x27;s response is often very textbook and ignores the reality on the ground ( corruption, non enforcement of laws and regulations, shakedown from local magnates &#x2F; police, etc).
  • seizethecheese16 hours ago
    It looks like it&#x27;s being served extremely fast via Openrouter right now <a href="https:&#x2F;&#x2F;openrouter.ai&#x2F;google&#x2F;gemini-4-neon" rel="nofollow">https:&#x2F;&#x2F;openrouter.ai&#x2F;google&#x2F;gemini-4-neon</a>. 127 tokens per second. This is way faster than Carbon at 46 tokens per second (<a href="https:&#x2F;&#x2F;openrouter.ai&#x2F;google&#x2F;gemini-4-carbon" rel="nofollow">https:&#x2F;&#x2F;openrouter.ai&#x2F;google&#x2F;gemini-4-carbon</a>). I&#x27;m a bit confused by this, since it seems like Neon is supposed to be their flagship model. Perhaps they&#x27;ve been throttling Carbon?<p>Carbon has a huge problem that it&#x27;s just way too slow, and outputs way too much per task (on real world tasks, it&#x27;s the worst at this <a href="https:&#x2F;&#x2F;micjm.com&#x2F;blog&#x2F;carbon-is-that-slow&#x2F;" rel="nofollow">https:&#x2F;&#x2F;micjm.com&#x2F;blog&#x2F;carbon-is-that-slow&#x2F;</a>).<p>Without the speed issue, Carbon would be a top model already. It&#x27;s top on our battle system built on opencode, if you exclude early votes: <a href="https:&#x2F;&#x2F;agentduel.com&#x2F;leaderboard?excludeEarlyVotes=true" rel="nofollow">https:&#x2F;&#x2F;agentduel.com&#x2F;leaderboard?excludeEarlyVotes=true</a>, though to be fair, small sample size so far.<p>Conclusion is that it&#x27;s looking good for Neon. Even if it suffers from similar problems as Carbon, if it&#x27;s even smarter and way faster, it should be a top model, maybe the best.<p>Edit: I was wrong about Neon being the larger model. I made this assumption because Neon has higher scores and they&#x27;re both &quot;v4&quot;.
    • WarmWash16 hours ago
      Neon is an Opus sized model<p>Carbon is a Fable sized model<p>Carbon seems to be a version of Argon that is incredibly spendy with thinking tokens.
      • seizethecheese16 hours ago
        Okay, that clears it up. Their naming convention is confusing because Neon has a larger atomic number than Carbon, so there&#x27;s just no pattern here at all.<p>Weird because Neon is outscoring Carbon in AA index, Terminal bench science, etc and they&#x27;re both named &quot;4&quot;. Should have released this as Neon 4.1.<p>With that clarification, I&#x27;m sure Neon has the same overthinking problem.
        • WarmWash16 hours ago
          Neon is weeks newer than Carbon, hence the improvements.<p>Google also has a rumored Krypton internal model, which is also Argon&#x2F;Carbon class. Neon might very well be distilled from Krypton, not Carbon.
          • seizethecheese16 hours ago
            It&#x27;s not weeks, Carbon was released on the 10th, 8 days ago.<p>&gt; Neon might very well be distilled from Krypton, not Carbon.<p>That would make a lot of sense, interesting!
            • WarmWash16 hours ago
              Right, but Google has had Carbon internally since early October IIRC.
          • desterothx15 hours ago
            in the model card they pretty much say it&#x27;s based on carbon
        • elxr11 hours ago
          AA index doesn&#x27;t tell you much. Some small models with low AA index are still incredibly useful because you know they can stay coherent and behave well over long tasks, and use your harness effectively despite the lower supposed &quot;intelligence&quot;.<p>Some giant models overengineer things and add redundant checks everywhere instead of simplifying overall design. In humans, that&#x27;d be a sign of low intelligence.
      • aenis16 hours ago
        [citation would be appreciated]
        • WarmWash16 hours ago
          Google (and the other labs) don&#x27;t give raw model sizes. It&#x27;s inferred from price&#x2F;speed&#x2F;benchmark&#x2F;quota. When you align those parameters, it&#x27;s clear what models are competing with what.<p>Also neon seems to be stepping on the toes of Carbon a bit, just like Opus 5.5 did to Fable.
          • seizethecheese16 hours ago
            Well, Anthropic named Opus 5.5 and Fable 5.1 then 5.2 to indicate that the smaller model was flying ahead.
            • WarmWash16 hours ago
              I&#x27;m almost positive google is purposely trying to breakaway from names describing sizes. If it works (read: helps balance demand), then the other labs will probably do it too.
              • syradar8 hours ago
                Well it’s making me not wanting to try their models even more.
              • desterothx15 hours ago
                it would be funny if they did some marketing around the hydrogen model (like as powerful as the h bomb) to try to push more non coders to use it for normal tasks
          • BoorishBears2 hours ago
            I mean Opus was Fable sized in the 4.5 days, and even larger in the 3.0 days<p>Google has the fuzziest lineup because of TPUs and their arbitrary pricing habits, Carbon and Argon almost seem same sized, Neon and Hydrogen are the only clearly smaller ones
    • nbardy6 hours ago
      Google has fully moved to rolling releases paradigm as their internal RSI starts to work, we&#x27;ll see more rapid releases that improves on all fronts(size and capacity) and it starts to confuse the model size paradigm.<p>You can often have a new set of RL envs or RL algos that train a smaller model past the capacity of the bigger model, especially with distillation involved and the tradeoffs of RL rollouts vs pretrain.
    • Melatonic16 hours ago
      I wonder if this is an experiment with their new partial chip design that&#x27;s vaguely similar to Taalas
      • seizethecheese16 hours ago
        It looks like it&#x27;s actually a smaller model than Carbon, just really confusing release.
    • elxr11 hours ago
      &gt; Carbon has a huge problem that it&#x27;s just way too slow<p>And makes shit up, and costs significantly more than Sol&#x2F;Opus on tasks requiring lots of decision making.
  • nicoburns17 hours ago
    Is there a way to determine &quot;model size class&quot; from Google&#x27;s naming scheme?
    • onlyrealcuzzo17 hours ago
      I <i>suspect</i> they may want to get away from the size class due to the fact that - at the pace they are launching new models - and the similarity of model sizes between Neon&#x2F;Carbon, Opus&#x2F;Fable, Sol&#x2F;Astra - the best model is going to be the most recently released one, not the biggest one.<p>I&#x27;m not sure it&#x27;s much better than 4.1 Flash vs 4 Pro -&gt; if 4.1 Flash is better temporarily when Pro is typically significantly better, people are going to be confused.<p>I&#x27;m not really sure how you get around this problem easily.<p>It&#x27;s not going to age well when the next Flash-lite model launches, but maybe they will call that `Gemini 4 &lt;Element&gt;-lite` - who knows...
      • nicoburns16 hours ago
        Interesting. You&#x27;re claiming that the Opus&#x2F;Sol&#x2F;Neon class and Fable&#x2F;Astra&#x2F;Carbon class models are similar sizes to each other? That would be surprising to me.<p>I&#x27;ve found that while Opus 5.5 is often pretty nice for practical work, Fable (even 5.0) is still significantly smarter on hard tasks and makes far fewer &quot;dumb mistakes&quot;
        • dannyw16 hours ago
          A Microsoft Azure page briefly claimed Sol runs with less looped transformer passes; and so you could speculate that Sol and Astra might share weights; but have different “loop” factors.<p>That would make a lot of sense from an inference optimisation perspective. The same prefill and decode machines can hold one set of weights, and serve both Astra and Sol requests.<p>At (multi-)terabyte-scale model weights, and when you have to offer not just the latest like Fable 5.2 but still serve 5.1, 5.0, etc; heck I think Opus 4.6 is still around; you really care about overall compute utilisation.<p>Idle GPUs are wasted GPUs; and even over the fastest interlinks you’re not hotswapping multi terabytes of weights.<p>With MoEs, there’s probably stuff you can play around with varying the number of active experts; or other clever optimisations.<p>Maybe Sonnet 5.5 could be just Opus 5.5 with a different router and less active params? Who knows; I’m just theorycrafting now :)
          • onlyrealcuzzo16 hours ago
            &gt; A Microsoft Azure page briefly claimed Sol runs with less looped transformer passes;<p>This doesn&#x27;t really make sense given how the models released.<p>OpenAI released Astra long before latent-space looping was a thing at the frontier space.<p>Then they released Sol 6, which was fast, clearly not looped. Then they release Sol 6.1 which was nearly as good as Astra, but way cheaper, and <i>WAY</i> slower -&gt; this would follow it just being a looped version of Sol 6.<p>It doesn&#x27;t seem likely that Astra 6.2 is just a looped Astra 6... It&#x27;s too fast, and they&#x27;d go bankrupt if they&#x27;re charging the same price... It <i>may</i> be a further looped version of Sol than Sol 6.2 - which would explain why it&#x27;s not as much better as we would&#x27;ve hoped.<p>It may also explain why their release is not going as well as they hoped. It&#x27;s hard to predict how your first looped models are going to fare.<p>If you look at how long it took Google to recover from their failed 3.5 Pro, it would be surprising if GPT recovered from their rumored failed run on Astra in under a month. That would give them reason to release a looped Sol and call it Astra.
            • iamdelirium16 hours ago
              Looped transformers have been known for years now. I don&#x27;t think it&#x27;s something new.
              • onlyrealcuzzo16 hours ago
                They are <i>JUST</i> arriving to frontier models.<p>Even in the open weight Chinese models as small as DeepSeek V4.1 Flash - nothing is looped.
                • refulgentis16 hours ago
                  I think they&#x27;re responding to the claim Astra can&#x27;t be looped, there&#x27;s clear and convincing research, as well as public comms, that it is. It&#x27;s unclear why Astra can&#x27;t be looped because looping is new, but, 6.1 Sol released a week later can be.
            • refulgentis16 hours ago
              Astra was the first shipping looper. Unconvincing evidence would be reporting x OpenAI&#x27;s quick reaction to it via a blog post from their chief research (it broke what&#x27;s considered a safety taboo). Convincing evidence would be research from ex. Neel Nanda on serial reasoning, recent update at <a href="https:&#x2F;&#x2F;x.com&#x2F;EpochAIResearch&#x2F;status&#x2F;2122762411287748696?a=J0EfVrGxNuIqrv1Fw1s&amp;s=20" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;EpochAIResearch&#x2F;status&#x2F;2122762411287748696?a=J...</a>
          • nicoburns16 hours ago
            I heard rumours that Sol 6.1 was cut down Astra because Sol 6.0 was received badly. But I thought that was the exception (and a result of Open AI&#x27;s actual Sol-class model being uncompetitive) rather than the rule.
            • willy_k16 hours ago
              That would line up - on release and for the few following weeks Sol 6.1 felt like Astra but maybe more task oriented, recently it’s been a noticeable downgrade in the level of specification needed to successfully set off.
        • onlyrealcuzzo16 hours ago
          Fable &amp; Opus, Astra &amp; Sol, and Carbon are all thought to be &gt;1T param models - AFAIK, most people think Opus &amp; Sol are ~5T param models and Astra, Fable are 10T+.<p>I think the rumor-mill consensus is that Google&#x27;s models are significantly smaller while attempting to swing at the same level - but the ratios AFAIK are thought to be around the same - i.e. Neon is <i>somewhere</i> around half as big as Carbon, big enough that a newer checkpoint can make it temporarily a little bit better.
          • ml-anon16 hours ago
            So many figures and ratios pulled from a single ass
        • weitendorf16 hours ago
          There&#x27;s only a limited amount of hardware for training and serving models at scale, and which form an upper bound for what model size&#x2F;inference can support.<p>A GB300 rack has 72x288GB=20TB VRAM. Do the math and 10T parameter models are just about the max size you can easily train and serve (at 1M context) without doing a lot of custom hardware development or limiting your scale.
        • andai16 hours ago
          <i>The Bitter Lesson - Richard Sutton (2019)</i><p><a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=19393432">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=19393432</a>
      • cl-coder2 hours ago
        &gt; Gemini 4 &lt;Element&gt;-lite<p>Surprised if it’s not going to be 4 electron&#x2F;proton.
    • XCSme17 hours ago
      <a href="https:&#x2F;&#x2F;blog.google&#x2F;innovation-and-ai&#x2F;models-and-research&#x2F;gemini-models&#x2F;introducing-gemini-4-neon&#x2F;#:~:text=Where%20Neon%20sits%20in%20the%20Gemini%204%20family" rel="nofollow">https:&#x2F;&#x2F;blog.google&#x2F;innovation-and-ai&#x2F;models-and-research&#x2F;ge...</a>
      • jeffbee17 hours ago
        You violated the only constraint the question contained.
        • ehsankia16 hours ago
          Honestly, the bigger question is, are they at some point start adding numbers (Carbon 2, Neon 1.5, Hydrogen+ ionized), or are they keep giving a new name to every single new model. Argon seems to have been basically &quot;replaced&quot; by Carbon.
          • kzrdude16 hours ago
            I hope they will continue to use numbers. Right now it&#x27;s &quot;Gemini 4 Hydrogen&quot; in the UI, that seems to be the way it will be used.
          • lukax16 hours ago
            Carbon is just Argon with extended thinking.
          • andai16 hours ago
            Subscribing for Gemini 19 Bismuth
            • grantsucceeded8 hours ago
              definitely the funnniest of all element names :)
        • XCSme16 hours ago
          Think of it like a riddle: given the order and naming scheme, OP has to imagine what Google thinks when naming the models...
        • antonvs7 hours ago
          Looking up the model name in a table is &quot;a way to determine &#x27;model size class&#x27; from Google&#x27;s naming scheme&quot;.
      • onlyj16 hours ago
        Why does this chart not include the other gemini 4 model, Argon? Also I am sure somebody thought about which element to name which model and it would be very helpful to me if they would explain their reasoning. OpenAI&#x27;s naming got kind of abstract with 5.6 but I could follow it: Luna(Moon) was smallest, Terra (Earth) was medium, Sol (Sun) was biggest. Astra kind of breaks from that but whatever. But like I would genuinely appreciation someone from google to explain the reasoning behind their element names. Like are Argon and Neon related in some way because they are both noble gases? Is hydrogen the smallest model because it is the lightest element? Carbon has a lower atomic number than neon but is more expensive and presumably bigger... maybe I am overthinking this.
        • thevillagechief16 hours ago
          Argon is already deprecated.
          • _delirium15 hours ago
            Inconsistently it seems. On the $20&#x2F;mo tier, Argon has been removed from the web interface and app, but is still available in Antigravity (while Carbon is only on the $100&#x2F;mo tier in both).
          • LewisOrton11 hours ago
            is it supposed to be replaced by neon?
          • jeffbee15 hours ago
            Uh yeah, is this the big news of this announcement? They are moving all the good coding models into the Ultra price tier.
            • Ternari14 hours ago
              Are they? Neon seems great and it’s available on the Pro tier.
              • jeffbee13 hours ago
                I&#x27;m going by the model labels in Gemini.
      • troupo16 hours ago
        Leave it to Google to create a card that goes from highest to lowest, and not the other way around
    • daemonologist16 hours ago
      I have to assume they&#x27;re messing with something other than size here - nobody would be <i>this</i> bad at naming different model sizes. Maybe sequence length or post-training strategy or sparseness or something else with the architecture.<p>(Although why they&#x27;d release all ~four is a mystery. I guess they&#x27;re having a hard time picking the &quot;best&quot; one and so are just putting them all out to let the public decide.)
    • ml-anon16 hours ago
      “Model size class” is a completely meaningless taxonomy.<p>It doesn’t correlate with intelligence, speed or price furthermore you have no way of telling what the model sizes of “like for like” named models from oai or Anthropic are.
      • NiloCK12 hours ago
        Shouldn&#x27;t it correlate pretty strongly with price? Like, model size is literally the amount of memory &#x2F; compute it takes to elect each token.<p>Agree there is no direct relation to quality - a Fable whose weights have been totally randomized would be the same size and cost the same amount to run.
        • ml-anon2 hours ago
          if you think companies are charging something which in any way corresponds to the cost per token I have an IPO to sell you.
      • sdenton416 hours ago
        &#x2F;me pours one out for neural scaling laws
        • ml-anon16 hours ago
          Did you read and somehow fail to understand the chinchilla paper?
          • sdenton413 hours ago
            Do you believe the amount of available training data has stopped growing?
    • DonsDiscountGas17 hours ago
      I assume atomic weight. Neon seems to be better than Carbon, also faster and cheaper (all judging by the link)
      • daemonologist16 hours ago
        What about Argon though?
        • breakwaterlabs16 hours ago
          Apparently even Google wants to pretend it doesn&#x27;t exist.
      • dudeinjapan11 hours ago
        Neon is more noble.
      • Tadpole918116 hours ago
        But Argon has a higher atomic weight than both, and sits below Carbon and above Hydrogen. Also, Carbon is rated at better at long-horizon tasks while Neon seems to be smarter at contained work. So they&#x27;re not strictly better&#x2F;worse.
      • ehsankia16 hours ago
        Great, consumers now need to first learn chemistry to be able to figure out which model to use!
        • accrual16 hours ago
          That&#x27;s my thought too. If one needs a lookup table to understand the product by name then it&#x27;s a product issue, not a consumer issue, even if everyone &quot;should&quot; have the first N elements memorized by atomic $property.
    • whimsicalism16 hours ago
      meant to be an opus&#x2F;sol competitor, looks like. not fable&#x2F;astra
    • scotty797 hours ago
      You can probably infer that form tokens per second.
    • moonu17 hours ago
      melting point&#x2F;boiling point
      • ahofmann16 hours ago
        Ugh, this sound so overly nerdy, that it could be very true. I do not follow or care about the Google LLM releases just because the naming scheme is so convoluted (and the constant churn of products around the llms). I learned the anthropic naming scheme, it is stable and easy to follow. And this is why Claude is my daily driver. I don&#x27;t like change, I like stable and predictable progress. Open AI naming scheme is kind of easy to learn (bigger planet = bigger model). But fuck Googles bullshit naming crap, seriously.
    • catigula17 hours ago
      If there is, it isn’t obvious, which makes it terrible.
    • drum5517 hours ago
      The developer guide lists<p><pre><code> Gemini 4 Neon (NEW!) Gemini 4 Hydrogen (NEW!) Gemini 4 Carbon (NEW!) Gemini 4 Argon (NEW! Gemini 3.8 Live Gemini 3.8 Flash (NEW!) Gemini 3.7 Flash Gemini 3.6 Flash Gemini 3.5 Flash </code></pre> I&#x27;ve no clue whatsoever how to read anything about the names of them and what they might imply you would use them for though. Within these most of them have LOW MEDIUM HIGH XHIGH, with the default being different for some of them, how anybody is possibly meant to understand this whole matrix of ambiguously named products is beyond me.
      • aenis16 hours ago
        Same here. First i thought it would be noble gasses. Then I thought it would be about the atomic weights. I know, I should just ask AI to figure it out for me, but its a bit humiliating.
        • drum5516 hours ago
          <p><pre><code> Hydrogen 1 Carbon 6 Neon 10 Argon 18 </code></pre> Maybe this implies that Neon is smaller than Argon in weight size, but it&#x27;s newer, but it&#x27;s also only a little bigger than Carbon? Do the Nobel gas models have more safeguards? Is Carbon 6x the size of Hydrogen and a third the size of Argon? When they release Xeon can we make compounds? Feels like some really cheap Da Vinci Code esque thing if we&#x27;re supposed to get some sort of meaning out of the names.
          • svnt16 hours ago
            From their Neon marketing:<p>&gt; In our early evaluations, it matched or exceeded Gemini 4 Carbon on overall quality at 3x the speed and 60% of the cost. We’re excited to bring these capabilities to Glean customers tackling multi-step work that demands accuracy, detail, and speed.<p>So possibly smaller than Carbon? Or some other axis that is making it faster and cheaper? No idea.
            • aenis16 hours ago
              Same. Running my benchmarks to verify their quality claims. Carbon performed very well on my private suite - spitting distance to Opus 5.5.
              • aenis11 hours ago
                ...first runs are in, and Neon is far below Carbon in terms of quality, but indeed much faster - ~4x faster, and 2x cheaper per task. Probably a good model for delegating to after key design work is done with something cleverer.
                • svnt9 hours ago
                  What category of work did it underperform on for you? I am seeing comparable outcomes in some technical document processing, but the cost is not lower for me.
                  • aenis6 hours ago
                    Gold plating for tasks where the agent is given the intention, the explicit ask to make the best impression, and ways of self evaluating. Much like openai&#x27;s agents pre 6.2 sol&#x2F;astra, it considers the task satisfactorily completed a bit earlier than anthropic models which seem to better guess what the specs left out. It is a useless feature for those who keep their llms at a short leash - my hobby benchmark aims to evaluate performance at under-specified tasks and, well, vibe coding as it was described a year ago.<p>It&#x27;s a good model for delegating work to, but not deciding the detailed scope. So, not fable class for me.
        • iainmerrick16 hours ago
          It could be worse, they could have named them after different Google chat products.
        • jimbokun16 hours ago
          How do you decide which model is best for telling you which model to use?
      • elxr11 hours ago
        Don&#x27;t use Argon or Carbon.<p>Hydrogen is the cheap one for straightforward tasks.<p>Neon is supposedly Opus-level, which is the exciting one if it actually works as well as they claim.<p>Obviously anything with a 3.x is not what you should reach for. This has always been the case on a first digit bump in any of these mainstream models.
      • prymitive16 hours ago
        It’s just a matter of time before we have<p>- Gemini (New)<p>- Gemini<p>- Gemini (original)
        • murderfs16 hours ago
          This is Google, it&#x27;d be:<p>- Gemini (Not ready yet)<p>- Gemini (Deprecated)
        • abixb16 hours ago
          Inb4 they switch it back to Bard.
        • accrual16 hours ago
          They could take a page from Nintendo: New Super Gemini Deluxe!
        • jimbokun16 hours ago
          Classic Gemini
      • kzrdude16 hours ago
        Hydrogen is the new flash, at least
    • Mogzol16 hours ago
      As far as I can tell it&#x27;s arbitrary and Google is just choosing element names they think sound good.<p>When Argon first came out it seemed like they could be going for noble gasses, like Helium (small), Neon (mid), Argon (big).<p>But then Carbon came out and it seemed like maybe they were going through the alphabet for model generations. Argon, Boron (internal), Carbon.<p>And now they&#x27;ve come out with Hydrogen and Neon, and there&#x27;s no logical way to order them. The actual model sizes are Hydrogen (small), Neon (mid), Carbon (big). This doesn&#x27;t match sorting by atomic number or alphabetically.
    • WarmWash17 hours ago
      They are almost certainly being ambiguous so that people rely on the model descriptors to pick which one to use.<p>I think there are a ton of people who automatically default to the largest model with the highest thinking for everything. Which historically made sense but not nearly as much anymore.<p>At this point a model like hydrogen can cover just about everything an average person is doing. You don&#x27;t need Carbon deploying an agent swarm pushing terabytes of parameters through HBM memory to help you plan your birthday party.
      • jeffbee16 hours ago
        Wouldn&#x27;t such users immediately run into quota walls?
        • WarmWash16 hours ago
          They do, and then they complain.<p>Worse, people burning dumb prompts on large models also has the net effect of shrinking quotas for everyone.
    • kphorn16 hours ago
      They use the very precise &quot;it sounds cool&quot;. Neon is 75% the price of Carbon but has 4 extra protons in the nucleus and is also a noble gas. Argon is a heavier noble gas at 16 protons.<p>element&#x2F;model - atomic number - cost H - 1 - $1.50 &amp; 7.50 per mtok C - 6 - $4 &amp; 20 per mtok Ne - 10 - $3 &amp; 15 per mtok Ar - 16 - $2 &amp; 10 per mtok<p>Claude - length of story but fable messes it up Haiku - 3 lines Sonnet - 14 lines Opus - a novel? Fable - 30 lines?? Mythos 15700 lines?<p>Luna - 28 days Sol - 365 days Astra - 5-6000 stars
    • adithyareddy16 hours ago
      (S) Gemini 4 Hydrogen - fastest - $1.50&#x2F;$7.50<p>(M) Gemini 4 Neon - second fastest - $3&#x2F;$15<p>(L) Gemini 4 Argon - second slowest - $2&#x2F;$10 (till Nov 30), $4&#x2F;$20 (&gt; Nov 30)<p>(XL) Gemini 4 Carbon - slowest - $2&#x2F;10 (till Nov 30), $4&#x2F;$20 (&gt; Nov 30)
      • WarmWash16 hours ago
        Argon and Carbon are the same class model. Carbon is just a newer rev of it.<p>There is a Helix model too (in the pipeline) which is the actual small size.
    • baoooooooooooo16 hours ago
      Everyone seems quite upset&#x2F;perplexed by the Gemini naming scheme but I quite like it. And hey, if you find yourself lost, the periodic table is probably the most ubiquitous ordered list in the world and serves to clear things up pretty fast. This will likely not be strictly true as models release over time, but is a good rough indication of what the team thinks we should expect of the models, in the same way the other labs run their naming.<p>I also like the personalities implied by these, with Anthropic saying “human literary achievement is neat”, OpenAI saying “space is neat” and Google being like “science is neat”.<p>Edit: I was coming back to say “I wish we could talk about stuff more substantive than the naming scheme” but (1) my comment may be objectively incorrect, and (2) I have to apologize that I haven’t helped the situation by… commenting on the naming scheme.
      • WarmWash16 hours ago
        They are not following the periodic table in any discernible way though.
        • dudeinhawaii16 hours ago
          They&#x27;re in alphabetical order, almost like Ubuntu release names. There was a leak about a Gemini Oxygen model, which would further support alphabetical order -- skipping names they don&#x27;t want to use.<p>So, names don&#x27;t align with capability but instead &quot;release freshness&quot;. This is unlike every other provider but it is what Google does (did?) with Android releases I believe (dessert names).
          • WarmWash14 hours ago
            They went form A to N in 4 releases. Initially maybe they planned to follow the alphabet, but I think that went out the window
      • georgemcbay16 hours ago
        &gt; And hey, if you find yourself lost, the periodic table is probably the most ubiquitous ordered list in the world and serves to clear things up pretty fast.<p>But the way google ranks the models by capability doesn&#x27;t follow periodic table order.<p>Hydrogen - 1<p>Carbon - 6<p>Neon - 10<p>Argon - 18<p>It could very well be based on their density (g&#x2F;cm^3 in standard conditions):<p>Hydrogen - 0.0000899<p>Neon - 0.000900<p>Argon - 0.001784<p>Carbon - 2.267<p>This would make more sense than assuming it is based on their periodic table order as it more accurately maps to how Google markets their capabilities, but is too esoteric to really be useful for clearing up confusion for normal users, IMO.
        • bertmuir6 hours ago
          I think it&#x27;s more vibes-based. The different periodic groups have different meaning.<p>Neon and Argon are descending down the noble gases, I&#x27;d expect their successors are Krypton, Xenon, Radon.<p>They skipped Helium and went straight to Neon, which makes me think they&#x27;ve reserved Helium as a lightweight element, and are going to follow the lightest elements for the cheapest models. Hydrogen, Helium, Lithium, and so on.<p>Carbon reads more as a &quot;foundation of life&quot; kind of metaphorical choice, and they have some choices in the periodic groups nearby to keep that sort of metaphor for a while. My guess is they move horizontally - Nitrogen, Oxygen - then on to Silicon when they hit something appropriately transformative with the model.<p>I&#x27;d expect the most experimental or fastest models to eventually release on highly reactive groups - Fluorine, Sodium, etc. - and that they&#x27;re leaving the metallic elements in the middle for differentiation in industry-specific fine-tuned models.
      • wat1000016 hours ago
        The periodic table isn&#x27;t really helping. Hydrogen is the lightest, that much makes sense. Beyond that it doesn&#x27;t seem to work. The elemental ordering is H C Ne Ar. The cost ordering is H Ne Ar C (with Ar and C being the same per token but I gather that C thinks more and thus costs more). The capability ordering seems to be H Ar C Ne. Argon is the highest on the table but is neither the most expensive nor the smartest.
      • pavo-etc12 hours ago
        Lmao no. The most ubiquitous ordered list is... numbers. Or the alphabet.
    • genxy17 hours ago
      I think you mean say nobel size, I&#x27;d wait for radon.
      • swasheck16 hours ago
        now we have to know our atomic weights to be able to choose which gemini model for the task, i suppose.
      • CamperBob216 hours ago
        Problem with Radon is that they will swap it out for Polonium without telling you, and then they will obsolete that in favor of Lead.
    • inquirerGeneral15 hours ago
      [dead]
  • onlyrealcuzzo17 hours ago
    This model feels like it might actually be as good as Opus 5.5.<p>This is probably the set of benchmarks most people care about: <a href="https:&#x2F;&#x2F;www.vals.ai&#x2F;comparison?m=claude-fable-5-2,cobalt-c-high-1p00,gpt-6.2-sol,gemini-4-carbon,claude-opus-5-5" rel="nofollow">https:&#x2F;&#x2F;www.vals.ai&#x2F;comparison?m=claude-fable-5-2,cobalt-c-h...</a><p>Overall, performs just under Fable (above Astra or Sol 6.2 or Carbon - slightly ahead of Opus 5.5).<p>Latency is by far the best for the level of quality.<p>It&#x27;s about 2x more expensive than Sol 6.2, but significantly better and faster. You get what you pay for.
    • SuperV123416 hours ago
      I am a Claude Ultra subscriber that&#x27;s very happy with Opus 5.5.<p>If I wanted to realistically try out Neon in some of my workflows without committing to a paid subscription before seeing if it can actually work as well as Opus, do I have any option, or do I just need to take the gamble?
      • user4392816 hours ago
        I was thinking of giving Gemini 4 Neon a try as a code reviewer instead of GPT-6.2 Sol, and found this in their FAQ:<p>&quot;Using third-party software, tools, or services to access Antigravity is a violation of our Terms of Service&quot;<p>and also: <a href="https:&#x2F;&#x2F;antigravity.google&#x2F;docs&#x2F;faq?can-i-use-antigravity-cli-headless-mode-in-third-party-coding-agents-or-automate-account-switching" rel="nofollow">https:&#x2F;&#x2F;antigravity.google&#x2F;docs&#x2F;faq?can-i-use-antigravity-cl...</a><p>I&#x27;d like to keep my Gmail account, so Google can keep their model and will not get my business.
        • WarmWash14 hours ago
          You can&#x27;t use subsidized tokens in 3rd party apps. Although they just suspend your Gemini account now, not your google account.<p>That being said, you can do whatever you want with API tokens.
          • user4392813 hours ago
            You can with Anthropic and OpenAI, using &#x27;claude -p&#x27; or &#x27;codex exec&#x27;.<p>Paying for Gemini at API rate is therefore out of the question for me.
            • p_l6 hours ago
              you can use agy in batch and&#x2F;or JSON mode - it&#x27;s what I do for few of my projects. You can also use &quot;agy acp&quot; to get ACP server now (used to be a separate binary)<p>The requirement is that you use antigravity harness to access subscription-based accounts, not that you can&#x27;t wrap it
              • user439285 hours ago
                Their official wording seems to be that it&#x27;s not supported:<p>&gt; Wrapping agy inside third-party coding agents or applications, or automating account rotation, is not supported under consumer plans.<p>At least they no longer ban the whole Google account, so I think interpreting that as a prohibition only for abuse would be reasonable.
            • joshuamorton6 hours ago
              That violates the ToS for both, and they absolutely will ban you (there&#x27;s tons of examples of people complaining about their cool tool to get Claude Code shimmed over an API not working&#x2F;getting them banned, same with OAI).
              • user439285 hours ago
                They won&#x27;t ban you for that.<p>I never had any issue using Claude for code review in Codex and vice versa.<p>I remember that half a year ago, Anthropic planned a change to make claude -p draw from a separate usage pool at API pricing, but after feedback that change was scrapped.<p>What people were banned for is using the subscription authentication in a third party tool, using many accounts, or reselling.<p>Invoking the official CLI from another agent for legitimate personal use during development should not be a problem.
        • fastest96314 hours ago
          I think that page is out of date. You can use it via T3 Code now and Google has said it&#x27;s okay. It&#x27;s via their official ACP-compatible endpoint that&#x27;s now provided by the antigravity-cli.
        • koolala11 hours ago
          You can install Codex in Antigravity its kinda fun running both.
        • what15 hours ago
          You don’t need to use the models through antigravity. You can get them through ai studio (or whatever Google calls it) and pay api pricing to use it via whatever harness you want.
      • dudus1 hour ago
        You want a Google AI Studio API key.
      • onlyrealcuzzo16 hours ago
        They say it&#x27;s going to be available for Pro users - so you <i>don&#x27;t</i> need a $200 account (Ultra). But, I don&#x27;t think you&#x27;re going to be able to try it unless you get at least a $20 account.<p>IIUC, you also don&#x27;t get to try Opus unless you have a subscription, right?
        • deaux13 hours ago
          Reminder that Google trains on your data if you&#x27;re using a consumer sub whereas Anthropic and OpenAI claim they don&#x27;t if you&#x27;ve got all relevant boxes turned off. I know everyone stopped caring about this at the start of the year and believes the latter two don&#x27;t honor it anyway, but it&#x27;s good to know.
      • layoric10 hours ago
        How do you find Fable 5.2? I tried it for a few days for coding, and I&#x27;m back on Opus 5.5, and even Deepseek 4.1 Flash for simple stuff. It was just going off on tangents way more for my workflow.
        • krisroadruck1 hour ago
          Glad I&#x27;m not the only one. I found Fable 5.2 to be a substantial step back vs Opus 5.5. It would happily burn a weeks worth of Ultra usage in a day to output an absolute slop pile of wrong assumptions and bad decisions. It just sucks for agentic use or even review work. Initial design back and forth is the only place I could get it to be useful. I&#x27;m starting to wonder if all the amazing examples anthropic gives for it only work if it has effectively an unlimited token budget so it can spin it&#x27;s wheels for however long it takes to eventually get a decent result.
      • DeathArrow7 hours ago
        You can use Neon on OpenRouter.
      • anentropic5 hours ago
        OpenRouter
      • desterothx15 hours ago
        openrouter or api directly? great question to use your claude sub to figure out easily :)
      • what15 hours ago
        You can always pay API pricing if you don’t want a subscription?
    • catigula17 hours ago
      0% chance this is better than any frontier OpenAI model.
      • HDThoreaun16 hours ago
        How can you be so confident about a model that was released today? I just dont get the lack of epistemic humility here
        • catigula16 hours ago
          &gt;how can one presume the sun will come up tomorrow merely because it has come up every day prior? have some epistemic humility! i am a very smart rationalist.<p>Edit: I tested the model and it found several mistakes Astra 6.2 made. It’s possible…
      • WarmWash14 hours ago
        I think that OAI likely has the strongest internal model, but I don&#x27;t think they can figure out how to ensure alignment on it
        • f13f1f1f113 hours ago
          I also have the hottest wife but no one can see her because people wouldn&#x27;t be able to constrain their lust. I need to hide her to maintain safety.
          • WarmWash13 hours ago
            However hot she is, she isn&#x27;t solving millennium problems.
  • bemtevi16 hours ago
    If one day, AI becomes provably good at all programming tasks, at working at mathematics better than all mathematicians, and at solving problems in almost all domains with maximal efficiency, I think it&#x27;d be a dystopia.<p>What I simply don&#x27;t get is why people want to push the boundaries of efficiency and knowledge. I used to think as a teenager that all knowledge was good but how good is knowledge when we can&#x27;t acquire wisdom with a comparable speed?<p>The greatest thing about the early internet and earlier science was that people shared and communicated with each other. Now they just share what they produce wihouth much humanity at all.<p>It seems to me that the creation of all these models has just become a race where no one really cares about the finish line and no one really understands why we&#x27;re doing it, except for the short-term trail of dollar bills on the race course.<p>For me, each new model release is a sad day for humanity.
    • bertmuir6 hours ago
      Because hundreds of millions of people live in abject poverty, and if we divvied our economic production up evenly today there&#x27;d only be ~$5,000 to ~$15,000 per capita at a US cost of living i.e. the global economy is far too small for everyone to have good lives.<p>The growth mindset isn&#x27;t really for you, if you&#x27;re already rich, and don&#x27;t want to be richer. It&#x27;s coming from the vast majority of the world, which is poor, and the minority that is rich but wants to get richer.
    • numlock866 hours ago
      &gt; If one day, AI becomes provably good at all programming tasks, at working at mathematics better than all mathematicians, and at solving problems in almost all domains with maximal efficiency, I think it&#x27;d be a dystopia.<p>Which of these are missing in your opinion? I&#x27;d argue we are there already. At least very very close, as &quot;with maximal efficiency&quot; is quite an arbitrary term.
      • bemtevi2 hours ago
        True. We&#x27;re getting close, though AI-assisted programming still needs human direction for complex projects and AI-assisted mathematics has trouble creating totally novel ideas. But yes, it&#x27;s close and a miserable situation to be in.
    • novaRom4 hours ago
      Don&#x27;t worry - AI regulations are coming soon.<p>A major point will be complete training-data transparency with independent expert review. In EU a statutory ban on collecting&#x2F;storing&#x2F;using users&#x27; data for training (including the &quot;de-identified data&quot; loophole), except voluntary explicit opt-in.<p>I wonder if a mandatory disclosure of training recipes paragraph will be included, likely in EU only (e.g. with tiered-access design to counter trade-secret objections) ?
    • simpsond11 hours ago
      We are already there, and yes, it often feels inhumane.
  • MisterBiggs14 hours ago
    I&#x27;ve been incredibly impressed with this model so far. Argon felt like a leap in presentation, the model writes really good documents and replies. Neon feels like another big leap in usability over that, the model seems pretty smart but its super eager and runs down tasks really well and efficiently. Definitely passes the vibe test
  • ByteAtATime16 hours ago
    This is an interesting model - I feel like they&#x27;re trying to avoid OpenAI&#x2F;Anthropic&#x27;s approach of various model tiers, and just keep climbing the Pareto graph (for lack of a better term) as fast as possible, both for cost and performance. I think I personally like the approach, despite the confusion it causes.
  • dannyw16 hours ago
    Major props to Google for announcing a model and releasing it on the same day.<p>Unfortunately, their recent change to allow me to opt out of AI training (which would also include connector data like Gmail, photos, etc) hasn’t rolled out to me yet[1]; so I won’t be upgrading my subscription; nor using the Gemini app until I can opt out.<p>[1]: <a href="https:&#x2F;&#x2F;arstechnica.com&#x2F;ai&#x2F;2026&#x2F;11&#x2F;google-will-let-you-opt-out-of-ai-training-in-gemini-without-losing-your-history&#x2F;" rel="nofollow">https:&#x2F;&#x2F;arstechnica.com&#x2F;ai&#x2F;2026&#x2F;11&#x2F;google-will-let-you-opt-o...</a>
    • sdvs9so15 hours ago
      I think that just changed as of a Nov 16th. You&#x27;re now able to opt-out your chats from model training while retaining chat history[1].<p>[1]: <a href="https:&#x2F;&#x2F;blog.google&#x2F;innovation-and-ai&#x2F;products&#x2F;gemini-app&#x2F;updated-controls&#x2F;" rel="nofollow">https:&#x2F;&#x2F;blog.google&#x2F;innovation-and-ai&#x2F;products&#x2F;gemini-app&#x2F;up...</a>
    • 5555watch16 hours ago
      &quot;Gemini History&quot; seems to only target Gemini App functionality, what about Antigravity chats?
      • nixosbestos15 hours ago
        well, there&#x27;s <a href="https:&#x2F;&#x2F;antigravity.google.com" rel="nofollow">https:&#x2F;&#x2F;antigravity.google.com</a> (not to be confused with <a href="https:&#x2F;&#x2F;antigravity.google" rel="nofollow">https:&#x2F;&#x2F;antigravity.google</a>)<p>Which ... doesn&#x27;t necessarily show what you want either. But you can access &quot;remote history&quot; in bottom left. Maybe, if it works for you.
    • deaux13 hours ago
      &gt; Currently, if you disable history, Google still retains that data for 72 hours (and 24 hours for the user). Once the updated page is live, this duration will change to two months.<p>So a regression then.
    • testtestmetest16 hours ago
      Oh know, they&#x27;re going to steal the code for you 3 MAU SaaS !!!<p>&#x2F;s not trying to be mean, its just weird hill to die on. Every company is stealing all your data in ways you unknowingly consent to at every second of the day.
  • jaykru16 hours ago
    Carbon 4 has been incredible for my use, tuning a local inference engine based on tinygrad running on Tenstorrent hardware. I had a barely functional prototype I built with Claude that petered out at an unusable level of performance. Astra and Sol didn&#x27;t make much progress when I pointed them at it. Carbon 4 more than 10x&#x27;d the prefill and doubled the decode performance on Qwen 3.8 27B without breaking quality or the generality of the stack. Very nice to see competition at the frontier again.
    • henrymerrilees11 hours ago
      This sounds like a blast!
      • jaykru8 hours ago
        So much fun! 3D printed a little rack for it too. I&#x27;m just dreading figuring out the household electrical stuff as I scale it up ;)
    • atomashevic16 hours ago
      Just curious, what TT hardware do you have? How many tps did you achieve with Qwen 3.8 27B?
      • jaykru15 hours ago
        I&#x27;m running 2x p150a&#x27;s right now. I have them hooked up to a mini-ITX PC with a single PCIe slot via a quad x1 switch (figuring out power to wire up the other two cards I own). My best numbers so far are around 2500t&#x2F;s prefill about 52t&#x2F;s decode with DFlash2. All of this is vibecoded besides the original design sketch which was maybe a 3 line description of what tweaks to make to tinygrad and which parts of the existing TT stack to use (almost nothing).
    • blueblisters10 hours ago
      A plausible explanation here is that Ant and OAI are deliberately sandbagging on ML &#x2F; GPU optimization tasks to preserve their lead while Google doesn’t give a fuck.<p>Argon is incredible at AI research and its responses are grokkable
      • jaykru10 hours ago
        It&#x27;s also really good at my fuzzier math&#x2F;research questions :shrug: more insightful than fable imo. The nerfing is real (indeed Ant at least admit to it) but I don&#x27;t really feel like it&#x27;s a useful fact to have in mind when evaluating the products I pay for.
  • stillpointlab16 hours ago
    I&#x27;m getting fatigued with the constant introduction of names and the impenetrability of mapping these brand new names to use cases.<p>I think internally, Google, Anthropic and OpenAI are thinking &quot;It&#x27;s so easy, you just use Neon for X, Carbon for Y&quot; - but most of the world is trying to map Sol, Astra, Neon, Opus, Fable, Mythos, Carbon, Haiku, etc.<p>I just hope they finally land on a naming scheme they are happy with and just stick with it. But I have my doubts. Perhaps they are all benefiting from this confusion.
    • aboardRat415 hours ago
      &gt;m getting fatigued with the constant introduction of names and the impenetrability of mapping these brand new names to use cases.<p>Because probably there is no system. They just train a model, and find out that it&#x27;s &quot;kinda good&quot;, and that&#x27;s it, they publish it. The models do not constitute even a partially ordered set, nobody can predict their ranking for your particular tasks.
      • motbus315 hours ago
        Like some vendors, making the name scheme incomprehensible is part of you not knowing which one is the best cost benefit
        • captainbland15 hours ago
          If you have to experiment to find out, you have to spend tokens
        • elxr11 hours ago
          Test it then.<p>Comprehend it using real use, or find someone you trust to use them in anger, don&#x27;t wait around for the labs to do everything for you. You&#x27;re the one with the leverage on where you send your money to.
          • motbus33 hours ago
            If you work for a company that requires several agreements for data and legal stuff it might take weeks to set each one up. It is not like you can simply use it everywhere. Apart from that, we always see opus dropping performance before another release, and openai started doing the same now. It is not like you have any guarantees on what you are buying
          • matkoniecz9 hours ago
            Maybe if you spend millions per year on tokens it makes sense to test all of them.<p>But at rate how many models appear it would cost more for me than potential savings.
          • jychang10 hours ago
            Nah, that&#x27;s like trying to test drive every single car on the market, in order to buy a car.<p>Just buy from companies with non shitty naming systems like Anthropic, and let the shitty names go bankrupt. Free market and all that.
        • paradox46014 hours ago
          Wonder when Sony is going to release a model, and pull it&#x27;s name out of a Scrabble word sack
          • mrandish13 hours ago
            True story about Sony naming: I was at the PC industry trade show where Sony launched the new VAIO computer in a big, beautiful booth at the front of the hall. Being a long-time Sony fan, near the end of the first day after the crowds had mostly cleared, I stopped by the Sony booth on my way out to see the new PC. I saw that the VAIO name also formed the computer&#x27;s distinctive ultra-modern logo featured in large bold text on each side of the PC&#x27;s futuristic case in a minimalist block font. But I was puzzled how a company with a history of so many great products with awesome names (WalkMan, DiscMan, PlayStation) had come up with such an awful name.<p>Since I was curious about the origin of the name, I wandered over to where three Sony employees were chatting passing time until the day ended. I learned all three were mid-level marketing staff based in Sony&#x27;s New Jersey US HQ. So I asked &quot;How do you pronounce the computer&#x27;s name?&quot; All three had different takes between various long vowel vs short vowel sounds.<p>Since they were all facing me, I also asked, &quot;Without turning around, how do you spell the name?&quot; They all paused, then laughed. They did manage to work that out after a few false starts but it wasn&#x27;t exactly smooth. As we chatted some more I asked who&#x27;d come up with such a unique name and learned Sony had hired a &quot;really cool&quot; high-end branding agency who specializes in coming up with great product names. The effort and expense in creating the name and logo had clearly been marketed to employees as &quot;cutting-edge cool.&quot; I warmly wished them the best with the new product before wandering out of the hall to dinner.
            • teshigahara13 hours ago
              None of it true of course. The VAIO name and logo was by Gotō Teiyū who had been working as a designer at Sony for 30 years at that point. Its pronounciation is quite unequivocal in Japanese as well (like bio).
              • mrandish12 hours ago
                Well, I guess the mid-level marketing people in Sony&#x27;s New Jersey office got some garbled version of what happened in Japan, because that&#x27;s what they told me at PC Expo &#x27;96 at the Javitz Center.<p>Regardless, the PC and it&#x27;s neat industrial design plus Sony&#x27;s brand and distribution were good enough to overcome a weak name. VAIO was an acronym for &quot;Video Audio Input Output&quot; (which Sony later changed to &quot;Visual Audio Intelligence Organizer&quot;, although I&#x27;m not sure why they thought that would be better).
                • teshigahara12 hours ago
                  Sure, I meant none of what they told you was true. I typically wouldn&#x27;t expect much better from sales people.
              • lejalv9 hours ago
                Yes, but, you know, the US is the center of the world
            • grantsucceeded9 hours ago
              i have to say, that ultraportable laptop was mind blowing to me at the time, maybe still.
      • floatingtorch12 hours ago
        There is definitely a system. They are training models of different sizes for different levels of inference efficiency, which maps to token pricing. The ordered set is the size of the weights&#x2F; computational complexity during inference.
        • aboardRat41 hour ago
          Complexity is objective, but weights don&#x27;t really tell you anything.<p>A model of the same size trained on random noise would still have the same size.
      • ijidak10 hours ago
        I feel Anthropic and OpenAI are actually doing a great job with their naming.<p>We intuitively understand that Luna is smaller than Terra, and Terra is smaller than Sol, and Sol is smaller than Astra.<p>Likewise, we intuitively understand that a haiku is shorter than a sonnet, and a sonnet is shorter than an opus, and an opus is shorter than a fable. Fable is probably the weakest of their names. Maybe they should have gone with Epic.<p>It&#x27;s Google that isn&#x27;t doing it right.<p>Even those who know physics have zero intuition regarding carbon, argon, and neon.<p>Hydrogen being the smallest makes sense.<p>But what property is the point of comparison for carbon, argon, and neon?<p>Valence shell? Atomic mass? I saw someone throw out density.<p>It provides no self-evident point of comparison.<p>Frankly, if they can&#x27;t even get naming right, they&#x27;re likely in trouble on the productization side.<p>It reminds me of Microsoft in the Zune era.<p>They just don&#x27;t have what it takes to win on the product side in this battle.
        • gfody9 hours ago
          Sol should be 400x larger than Luna
        • wpm9 hours ago
          Intuitive would be naming them &quot;small medium large&quot; or &quot;good, better, best&quot;, not depending on the customer to know Latin or the periodic table or fucking whatever.<p>Hell, steal the terrible muddy naming from Apple and it would be better. Claude, Claude Pro, Claude Max. GPT-5.6, GPT-Pro-5.6, GPT-Max-5.6. Whatever! Tiering is good and fine, just make it clear.
        • Daneel_9 hours ago
          Huh, I guess there is an order. Definitely not &#x27;intuitive&#x27; unless you actively try to reason about the names. I&#x27;d always assumed they were basically code names that had no bearing on size or performance.<p>I only understand the Luna&#x2F;Terra&#x2F;Sol&#x2F;Astra order though. I couldn&#x27;t have told you that a sonnet or an opus or a fable had any sort of sizing guidance. English go brrr.
        • grantsucceeded8 hours ago
          sorry, but it might as well be small, medium, large. oh, and maybe supersize.<p>its all so one dimensional, just call it moon, earth, sun and stars cause that will make us all warm and fuzzy.<p>oh, no wait, it&#x27;ll be way more informative and clear to call it terra,luna,sol and astra. Yeah, thats way more genuine caus it makes me feel smart for figuring out it means small, medium, large and supersize.
      • ijidak10 hours ago
        I feel Anthropic and OpenAI are actually doing a great job with their naming.<p>We intuitively understand that Luna is smaller than Terra, and Terra is smaller than Sol, and Sol is smaller than Astra.<p>Likewise, we intuitively understand that a haiku is shorter than a sonnet, and a sonnet is shorter than an opus, and an opus is shorter than a fable. Fable is probably the weakest of their names. Maybe they should have gone with Epic.<p>It&#x27;s Google that isn&#x27;t doing it right.<p>Even those who know physics have zero intuition regarding carbon, argon, and neon.<p>Hydrogen being the smallest makes sense.<p>But what property is the point of comparison for carbon, argon, and neon?<p>Valence shell? Atomic mass? I saw someone throw out density.<p>It provides no self-evident point of comparison.<p>Frankly, if they can&#x27;t even get naming right, they&#x27;re likely in trouble on the productization side.<p>Their fumbles remind me of Microsoft in the Zune era.<p>They just don&#x27;t have what it takes to take first or second place in this revolution.<p>But they&#x27;ll likely do fine indirectly via TPUs, the Google homepage, Gmail, etc. Just as Microsoft was okay because of their legacy dominance.
        • vojvod6 hours ago
          &gt; We intuitively understand that Luna is smaller than Terra, and Terra is smaller than Sol, and Sol is smaller than Astra.<p>Sol being smaller than Astra isn&#x27;t intuitive unless you think of it in terms of distance, which would imply Terra -&gt; Luna -&gt; Sol -&gt; Astra.<p>EDIT: Apologies, elsewhere someone pointed out that astra means stars plural rather than singular which I didn&#x27;t know so that makes more sense.
    • josephg12 hours ago
      With anthropic at least they have a naming system based on the size of the model, with literature analogies:<p>Haiku &lt; Sonnet &lt; Opus &lt; Mythos&#x2F;fable<p>I think openai has a similar system with astral bodies:<p>Luna &lt; Terra &lt; Sol &lt; Astra
      • mlinsey12 hours ago
        I like both naming schemes...but the thing that always confuses me is, aren&#x27;t fables pretty short? The Tortoise and the Hare is a lot shorter than say, Anna Karenina or other opuses...<p>Although, maybe it&#x27;s called Fable because it&#x27;s trying to bluntly convey strong moral lessons, hence the strong safety classifiers...
        • nimbleal9 hours ago
          I find Open AIs system confusing also because my mental metaphor for their names is a pre Copernican cosmos so it should go terra &gt; luna not the other way around
        • XajniN9 hours ago
          [dead]
        • amirhirsch11 hours ago
          One thing is for certain: cache invalidation is decidedly the second hardest problem in computer science
        • mattnewton11 hours ago
          It’s Fable because they used up the name Mythos internally already for the model class.
          • KingMob10 hours ago
            They could have used Epic, but that&#x27;s a heavily-overloaded term in software. Maybe Saga?
            • fastball10 hours ago
              Saga would&#x27;ve been good. Or they should&#x27;ve just released neutered Mythos to the public as Mythos instead of Fable, and then renamed Mythos to be like mythos-unlobotomized or whatever.
            • cozzyd10 hours ago
              Tome?
      • emaro12 hours ago
        I like them too but it took a while for me to land, tbh. Mistral does the best job IMO: Mistral Small&#x2F;Medium&#x2F;Large. Slap a number on for the generation.
      • rafaelc12 hours ago
        This doesn’t land for the vast majority of people. Maybe 1 &lt; 2 &lt; 3 &lt; 4 would work better.
        • mlinsey12 hours ago
          It makes sense if you understand that the size of a model is something important and meaningfully different than just how advanced the model is.<p>You could argue that&#x27;s a detail the user shouldn&#x27;t have to concern themselves with, but I think it&#x27;s appropriate to expect them to learn that and pick names which put that concept front and center. The various &quot;model router&quot; approaches for consumer tools like ChatGPT have been finicky and inappropriate for professional tools we spend a lot of money on.
        • Aeolun12 hours ago
          I’m not sure the vast majority of their audience doesn’t get it. I feel like it’s arguably clearer than 1-4 would be.
          • Daneel_9 hours ago
            I strongly disagree. The order of numbers are universally understood since they&#x27;re taught since childhood. The arbitrary order of literature terms and old-timey astrological names are not.<p>Better still is simply small&#x2F;medium&#x2F;large&#x2F;extra large, or low&#x2F;medium&#x2F;high&#x2F;maximum.
            • tdhz779 hours ago
              Or perhaps Size, version and year<p>Claude L126 Claude L226 Claude L327<p>Now we know which model is the latest and which version we are on in that year. Now give it a cute name for your blog. And when people use Intrepid, we will all be confused all over again.<p>Claude S126-Intrepid Claude S226-Jaunty Claude S327-Maverick
        • KylerAce12 hours ago
          But the number is for the model iteration not strength within a generation.
        • utopcell12 hours ago
          So Opus 5.5 would be &quot;3 5.5&quot;? I don&#x27;t think that&#x27;s better.
          • mrheosuper12 hours ago
            How about &quot;Claude 5 0505&quot; ? First number is tier(5 = opus, 1 is haiku, 3 is sonnet), next 4 numbers are version(2 for major, 2 for minor)
        • npunt12 hours ago
          &quot;What&#x27;s the #1 Claude model?&quot;<p>&quot;Number four, obviously!&quot;
          • bloppe12 hours ago
            duh, it&#x27;s number Astra
        • elxr11 hours ago
          Maybe if you don&#x27;t know a damn thing about branding. Imagine saying GPT-6.2-3<p>???
      • snthpy11 hours ago
        Exactly, so even if you don&#x27;t remember you can deduce the order. I just looked up the atomic weights of Neon and Argon and apart from Hydrogen the Google models are all in the wrong order. Unless I&#x27;m supposed to be using density?<p>Edit: Density seems to track, H &lt; Ne &lt; Ar &lt; C
        • snthpy11 hours ago
          Still seems the wrong way around from an antigravity perspective. Hydrogen should be the strongest and elevate you the most.
      • wejick10 hours ago
        Anthropic doing a good job sticking on the same naming from the beginning.<p>OpenAI&#x27;s is still confusing for me tbh, but Google&#x27;s is on different level!
      • okrad11 hours ago
        Google was close with atomic numbers of elements but Carbon bucks the trend.<p>Hydrogen 1 &lt; Neon 10 &lt; Argon 18 &lt; Carbon 6
      • mattlondon6 hours ago
        That only helps if you know the ordering of those things.<p>I certainly had no idea on the relative sizes of e.g. Sonnet or Fable. Who knows this kinda of english literature trivia?!<p>Likewise OpenAI I still don&#x27;t understand at all - isn&#x27;t Astra another word for Moon? So we&#x27;ve got Luna which is I don&#x27;t know what, Terra is something to do with clay I presume given terracotta, Sol I think is a beer from Mexico, and Astra the moon. What the fuck? I presume there is actually some sort of scientific or latin connection, but I have no fucking idea.<p>Maybe these things make sense to the PhD-laden halls of frontier labs but to average joes they are meaningless.
    • mstibbard16 hours ago
      I promise you the confusion is felt inside these companies too.
      • jtfrench14 hours ago
        This sounds highly likely. Imagine trying to name and market a bag of weights, and arguing with a straight face why it’s better than another bag of weights.
    • pavo-etc15 hours ago
      I maintain a table for all labs. each row is S &#x2F; M &#x2F; L &#x2F; XL (maybe 2XL 3XL etc in future) and each column is the lab. The only way i can contextualise any of this.
      • newman31414 hours ago
        Link to table please?
        • HJain1310 hours ago
          Not OP but I tried here (probably still wrong lol) <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=50556084">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=50556084</a>
        • robofanatic13 hours ago
          you can try asking your favorite LLM to make it and see how it goes
    • mrandish13 hours ago
      After looking over the landing page, my first thought was also about whether I wanted to invest the time to figure out where this new model fit in the relative hierarchies of comparable product stacks. Ultimately I decided &quot;Nah&quot;, and aborted to the comments.
      • oniony13 hours ago
        They have some interesting graphs linked. From these I learnt I probably should be using GPT Sol for its bang to buck.
    • crystaln13 hours ago
      Yeah when I talk to people who don’t follow ai closely they are very confused, particularly about Claude where you have to pick a model by an arbitrary name. At least ChatGPT has a slider for model strength
      • f13f1f1f113 hours ago
        Anthropic has at least been consistent these other ones change it every fucking week. I honestly just think of models in terms of haiku&#x2F;sonnet&#x2F;opus level now because the other ones are completely inconsistent.
        • piskov13 hours ago
          The issue with anthropic is a fable is smaller than an opus<p>So they fucked up naming as well in this regard.
          • tintor13 hours ago
            And you believe opus model has less parameters than fable?
            • vbezhenar12 hours ago
              There&#x27;s no &quot;Fable model&quot;. There&#x27;s Mythos. And it&#x27;s certainly larger than Opus.<p>Fable is Mythos with safeguards, basically.<p>In fact, Mythos also includes safeguards, so at this point it&#x27;s not really clear what&#x27;s the point. You can apply for security program and there are three levels of access to the Mythos. I guess now it makes sense to just drop &quot;Fable&quot; name and go with Mythos which will just be implicitly restricted according to your access level.<p>Or maybe they&#x27;ll realize that there are abliterated openweights models, and restricting Anthropic models does more harm than good. I literally have to use other service for security audit of my software, because Fable rejected it.
            • apetresc13 hours ago
              Yes, almost certainly.
            • slowin13 hours ago
              No, they mean the models are named after a form of writing where the longer form = a bigger model. Except an opus is longer than a fable.
          • MikeTheGreat13 hours ago
            To be fair, I think they wanted to name it &quot;Mythos&quot;, not &quot;Fable&quot;.<p>Then the Trump administration panic-banned Mythos and Anthropic smartly decided to maybe call it something else.
            • trollbridge10 hours ago
              Well, they ran around boasting about how dangerous Mythos was. Maybe they should call it Ragnarok
      • nimchimpsky9 hours ago
        [dead]
    • overleaf12 hours ago
      At work at Giant Tech Company, someone has started publishing a &quot;WTF&quot; guide every week that includes the decoder ring for model names vs. their capabilities. I just have my agent read it every week and update all my skills and rules appropriately to figure out which model to use for what.
      • ehnto9 hours ago
        Wouldn&#x27;t that mean your skills and rule outcomes differ every week?
    • SamuelAdams13 hours ago
      They really need an industry term for each model. Think like mobile, tablet, laptop, desktop. Apple, Lenovo, HP, Google, and others all have their own versions of these, but in general the public has some idea of common use cases that each product will satisfy without delving into the details very much.
      • 0x6942013 hours ago
        to be fair we have certain semi-common names like &quot;flash&quot; and &quot;small&quot;, even though those tend to be relative to the size of their flagship model, which makes the terms somewhat meaningless out of context... but at the very least those do help when you&#x27;re familiar with the given company.
    • Jayakumark11 hours ago
      They should standardize on naming like Medals for all models.<p>Medal - Anthropic Claude, OpenAI, Google Gemini , Meta Muse<p>Bronze - Haiku , Luna , Hydrogen, Ember<p>Silver - Sonnet , Terra, Argon (Now defunct) , Spark<p>Gold - Opus , Sol, Neon, Wonder<p>Platinum - Mythos&#x2F;Fable, Astra , Carbon, NA<p>Diamond - ?<p>One flaw is people using it will always try to stay on gold or above - in their mind, gold might be better so they may spend more when it’s not needed.<p>Silver is all you need now for most of coding work.<p>Edit: updated argon and neon categories on table above as Neon replaced Argon
      • danpalmer11 hours ago
        The Gemini models here are not quite in the right places. Hydrogen benchmarks at a Silver, Neon benchmarks at a Gold&#x2F;Platinum depending on what you&#x27;re doing. Argon benchmarks at Silver&#x2F;Gold depending on what you&#x27;re doing. Neon is very much above Argon.<p>I&#x27;m not at all familiar with the Meta models, are they correct, or just put in place based on market positioning?
      • Ethan_Mick11 hours ago
        This was actually very helpful, thank you! I need this. These tiers tell me exactly how to compare models and where to use them.
      • davidd_100411 hours ago
        Carbon&#x27;s price and performance are much closer to Opus than Fable. And Neon is really supposed to supersede Argon (in that same tier). That&#x27;s what makes it so confusing.
    • dimavolo10 hours ago
      I’d like to see them use Apple’s naming scheme:<p>Gemini Mini, Gemini, Gemini Pro<p>GPT Mini, GPT, GPT Pro, etc<p>And then just increase the version<p>GPT Mini 6.2 = Luna 6.2<p>Gemini Mini 5.1 = Gemini Neon<p>Etc<p>Maybe add a 4th&#x2F;5th one in there sometimes, eg to distinguish Gemini hydrogen from Gemini Neon, like “Mini” vs “Air” does
      • jezzamon10 hours ago
        They literally had this (flash, pro, etc) and then decided to discard it for some reason...<p>At Google we make memes about how the naming system is incomprehensible
    • rtkwe13 hours ago
      Yeah I don&#x27;t even try, whatever Copilot sends it to seems to work well enough for me at the moment.
    • pizzafeelsright12 hours ago
      I deal with this at a global level.<p>There isn&#x27;t yet a single good method. We&#x27;re getting close to unifying the naming practice.<p>The routing and harnesses are still not adapting fast enough but when they catch up the names won&#x27;t matter.
    • minimumvector12 hours ago
      What I can&#x27;t understand is how analogous in size these models are to each other. Is Fable roughly equivalent in size to Astra? Opus roughly equivalent in size to Sol? Is Neon more similar to Fable or Opus?
    • elxr11 hours ago
      Carbon&#x2F;Argon&#x2F;Hydrogen - Just skip. Simple.<p>Neon - Probably decent based on twitter impressions and benchmarks.<p>For anthropic, just listen to anybody that heavily codes agentically, it should be clear which model is the best for most typical tasks (Opus).<p>OpenAI though, it&#x27;s definitely less clear right now. I haven&#x27;t been enjoying either 6.2 astra or 6.2 sol, and they&#x27;re definitely occasionally bone-headed in a way I haven&#x27;t seen before on a frontier model this late into the game. If the name is the biggest of your concerns, you&#x27;re not doing enough work.
    • floatingtorch12 hours ago
      I wish they would all just use huge, big, small or something like a clear ordering. Similar reasoning levels which are very intuitive.
      • MrDunham10 hours ago
        Agreed... though at some point I wonder if they paint themself into a corner with that...<p>&quot;Even Huger&quot;, &quot;Most Hugest&quot;, and &quot;Most-Most-Hugest-v2_Final_Final&quot; feels like a real possibility.<p>Fable is a good example. &quot;It&#x27;s Mythos (real big)... but Mythos-light, so we&#x27;ll call it Fable&quot;. Which is smaller than an Opus.
    • robofanatic13 hours ago
      reminded me of Android dessert version names Cupcake, Donut, Gingerbread, Ice cream sandwich….
      • SoMomentary13 hours ago
        IIRC those at least did us the courtesy of proceeding alphabetically.
    • B-Con14 hours ago
      I&#x27;m <i>terrible</i> with names, so I&#x27;m in the same boat. There&#x27;s a lot of marketing going on.<p>But the naming isn&#x27;t completely random, here&#x27;s what I use to make the ordered list in my head:<p>Claude is named after creative works, and uses bigger works for more powerful models. In order of strength:<p>* Haiku - short poem (we all know this one)<p>* Sonnet - &quot;medium&quot; length poem (many of us know this one)<p>* Opus - It&#x27;s the singular form of &quot;opera&quot;, so it&#x27;s like a stand-alone segment of a larger body of work. (Nobody knew this one, but not hard to remember, and know you know that &quot;opera&quot; is actually plural! The description sounds bigger than a sonnet, but it&#x27;s not a strict ordering.)<p>* Fable - a story (this one doesn&#x27;t seem obviously bigger than an opus, but I guess an opus is one part and a fable is a front-to-end story, it works)<p>* Mythos - another type of story, which works and it&#x27;s basically Fable but Anthropic&#x27;s internal-only version<p>Gemini, as of version 4, started doing something with atoms:<p>* Hydrogen - smallest periodic element, the most common gas in outer space<p>* Neon - a bigger element, less common in outer space<p>* Argon - yet bigger element, yet less common in outer space (as best I can tell, I am not an astronomer)<p>* Carbon - this breaks that pattern, its a smaller element than the first three, and not a gas... but critical for life on this planet? And it sounds like a longer version of Argon. So I can remember it, but I&#x27;m not sure what the intended pattern was.<p>The model ordering is not strictly upgrades, but AFAIK they are intended to be steps up in terms of (claimed) complexity and reasoning power. But since it&#x27;s an evolving field, the specifics of how cost, speed, and accuracy combine can make it hard to know how to order a given model for your use case. (i.e. cheap &amp; fast with a harness that makes more calls, or expensive and slow but better at one-shot).<p>Deepseek&#x27;s naming is more organized and literal, but you still have to remember the prefixes for V = version (plain) and R = Reasoning, and that they have different specialized models that aren&#x27;t intended to be compared to each other.
      • adrianmonk12 hours ago
        I&#x27;m not sure if it&#x27;s intentional, but the Google AI code names are all elements used in producing light.<p>* Neon and Argon: noble gases commonly used in gas-discharge lamps[1].<p>* Carbon: used in arc lamps[2] (the first practical form of artificial electric light).<p>* Hydrogen: the sun (abundant light, available to everyone).<p>The symbolism of light would make sense for AI. Shedding light on a subject, enlightenment, a light bulb moment, etc.<p>If this is right, then maybe we&#x27;ll see krypton, xenon, mercury, sodium, or tungsten next.<p>---<p>[1] <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Gas-discharge_lamp" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Gas-discharge_lamp</a><p>[2] <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Arc_lamp" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Arc_lamp</a>
      • surround13 hours ago
        Basically every lab has a model that doesn&#x27;t fit their naming scheme 100%.<p>OpenAI&#x27;s naming scheme is in Latin:<p>* Luna - Moon (smallest)<p>* Terra - Earth (medium)<p>* Sol - Sun (large)<p>* Astra - Star (could be larger than the sun I guess)<p>Meta&#x27;s Muse:<p>* Ember - (smallest)<p>* Spark - (medium)<p>* Wonder - (large)<p>A spark is the beginning of an idea whereas wonder is associated deep thought. But an ember and a spark are both small pieces of fire, so the ordering is a little muddy.<p>Google&#x27;s naming scheme has the least clear pattern of any lab.
        • mmoskal12 hours ago
          Astra is actually stars plural. One is &quot;astrum&quot; apparently.
        • bee_rider12 hours ago
          I wonder if it was supposed to grow in fire size. Ember and spark fit the pattern and there’s plenty of room to grow. But eventually they’d start reaching sizes related to catastrophes.
          • Aeolun12 hours ago
            Muse Wildfire, Muse Conflagration
        • tintor13 hours ago
          Yes, it is Astra because &#x27;GPT 6 Black Hole&#x27; didn&#x27;t get approved by marketing department.
          • zzzoom12 hours ago
            And people don&#x27;t know Sagittarius A*
          • proton_913 hours ago
            haha! this made me chuckle, perhaps they release black hole when they can claim to have achieved singularity
      • throwaway47382514 hours ago
        Speaking of Opus, many have likely heard of the expression &quot;magnum opus&quot; (Latin for great work). That&#x27;s probably easier to remember than the singular form of opera.
      • fimi14 hours ago
        At first I thought that has lexicographic order, like Argon is small, Carbon is big, and something between.. but then Hydrogen, and Neon, whatever
      • LoganDark13 hours ago
        Mythos and Fable are identical now except for &quot;vetted organizations&quot;. I&#x27;m not even sure why they&#x27;re still offered separately in the CVP.
    • crazygringo14 hours ago
      Seriously. It&#x27;s annoying enough to keep track of the names of which models mean what just within Claude Code. But they don&#x27;t change <i>that</i> often, so I can deal.<p>But trying to keep track of the model names in all 3 companies? Utter madness.<p>We need something better. All I really care about most of the time is whether I&#x27;m using the cheapo, average, or expensive model. I don&#x27;t really care about which actual model is occupying which actual spot at which time with which cutesy name. I just know I have a budget, and so I allocate the three levels according to the type of work I&#x27;m doing.
    • whoanow15 hours ago
      [flagged]
    • nl12 hours ago
      I think vendors assume their names schemes are obvious.<p>Anthropic: &quot;We&#x27;ll name after the size of literature style! An opus is <i>obviously</i> bigger than a sonnet!&quot;<p>Much of the world: &quot;What is this word <i>Opus</i>?&quot;<p>OpenAI: &quot;We&#x27;ll use the Latin names for astronomical bodies. <i>Everyone</i> knows that terra = earth and luna = the moon and that the earth is bigger than the moon.&quot;<p>Much of the world: &quot;Does terra mean it is terrified? Oh well at least it is better than Gpt 4o vs Gpt o3&quot;<p>Google: &quot;We&#x27;ll name it Bard! Oh, that was bad. Lets name it Gemini because that sounds cool. Oh we need different models, ok we&#x27;ll have <i>Pro</i> for the big expensive on, and <i>Flash</i> for the fast one. But now we want a fast large one and also we have mid sized models. Ok, lets name them after elements, but not ones that anyone knows anything about so now we can confuse them as to if Neon or Carbon or Argon is better. But Gemini Hydrogen is the cheap fast one&quot;<p>Everyone: &quot;Ok, Gemini Hydrogen is cheap. But this is too confusing, I&#x27;ll just use Claude&quot;.
    • asaddhamani10 hours ago
      We had names like nano, mini, flash, flash-lite, pro, etc. These were easy to understand. I have no idea what’s going on with this helium hydrogen carbon germanium nonsense. Do they want me to take out a copy of the periodic table to understand how the models compare? Is that even how it works? Flash lite, flash, pro wasn’t broken. Why did they “fix” it?
    • datadrivenangel13 hours ago
      Poob. Poob has it for you.<p>0 - <a href="https:&#x2F;&#x2F;knowyourmeme.com&#x2F;memes&#x2F;poob-has-it-for-you" rel="nofollow">https:&#x2F;&#x2F;knowyourmeme.com&#x2F;memes&#x2F;poob-has-it-for-you</a>
    • willis93616 hours ago
      Maybe it&#x27;s the same marketing direction that landed starbucks&#x27; drink size names.
      • virtue316 hours ago
        Venti is just Italian for 20 for 20oz<p>Yeah grande(big?) and tall are whatever to your point
      • colejohnson6616 hours ago
        At least Starbucks doesn&#x27;t change their drink size names every few months.
    • eminence3215 hours ago
      It doesn&#x27;t help that the blog post opens with a long description of all the benchmarks that the model excels on, but very little info about how I should pick what model to use.<p>I do note that gemini.google.com has replaced Argon with Neon, so I guess maybe that&#x27;s a clue?<p>My personal approach is to not really care at all. These specific models aren&#x27;t going to be here in 6 months, so learning the intricates of these specific models has no benefit for me. I&#x27;m just going to mostly rely on whatever one is the default
    • com2kid11 hours ago
      If you ask Gemini what it thinks about Gemini model names even it&#x27;ll agree it is a bad product branding strategy.<p>So, there is that...<p>Luna, Tera, Sol, that makes sense. Size. Yay. Big thing better.
    • Art968111 hours ago
      The naming scheme got you to take precious time and talk about it. All marketing is good marketing.<p>I still agree with you though.
    • ionwake14 hours ago
      Just use animals guys. Predator prey for thinking, nimble more reactionary etc.<p>Expert models: Lion, Shark, Whale<p>Flash models: Swift, Falcon, Robin<p>Though I still think it shoudl jsut be : Expert v13 , Fast v3
      • nimbleal9 hours ago
        From a user’s point of view I think it really comes down to the old “fast cheap powerful — you can only pick two”. I’d love<p>Fast and cheap Fast and powerful Cheap and powerful Kinda all three but not really<p>Not very catchy names though
    • quixoticelixer-14 hours ago
      It&#x27;s possible they are just making it confusing as possible so everyone switches to the orchestrator but I bet they are just as confused as the rest of us.
    • pizzathyme16 hours ago
      I thought the higher periodic table numbers meant a more powerful capable model. But not even that appears to be the case
    • rao-v13 hours ago
      Atleast there is some hope we&#x27;ll get Gemini 5.3 Arsenic one day
      • tintor13 hours ago
        Gemini Lead, Gemini Uranium, Gemini Plutonium
    • coolestguy11 hours ago
      It&#x27;s completely ridiculous when not even tech people can keep up - how can a normal person?? Just name them, small&#x2F;medium&#x2F;large then a revision number
    • j9feng14 hours ago
      agreed! Apparently their AI is not intelligent enough to pick proper name schemes for itself.
    • paulddraper14 hours ago
      What do Corolla, Camry, Crown mean?<p>While cutesy, most current models are quasi-semantic progressions.<p>Claude (length, or cultural durability): Haiku, Sonnet, Opus, Fable<p>GPT (altitude): Terra, Luna, Sol, Astra<p>---<p>Gemini unfortunately has no semantic key Hydrogen, Carbon, Argon (superseded by Carbon), Neon.
      • motionlessveloc14 hours ago
        Corolla = Corona = Latin for &quot;Crown&quot;<p>Camry = Kammuri = Japanese for &quot;Crown&quot;<p>Crown = English for &quot;Crown&quot;
        • redwall_hp12 hours ago
          There was also a Corona until 2001, and briefly a Tiara in the 60s.
        • paulddraper13 hours ago
          That makes sense why Toyota replaced “Avalon” with “Crown”.
      • stillpointlab12 hours ago
        I thought about this. Does that mean renting a model is like leasing a car?<p>Anthropic is the new Mercedes? OpenAI BMW?
      • j0s412 hours ago
        You have the GPT ordering wrong though, as Terra is generally more capable than Luna
      • happyopossum13 hours ago
        &gt; What do Corolla, Camry, Crown mean?<p>This wouldn&#x27;t be a conversation if the big 3 didn&#x27;t change their model names multiple times per year. Or conversely - do you think people would be fine with Toyota renaming the Camry every 6-9 months?
    • closetheloopdev15 hours ago
      Maybe the trick to working with these RaNd names is for us to tell Claude Fable&#x2F;Opus or GPT Astra&#x2F;Sol: &quot;go find the Gemini model that is as smart as you and discuss XYZ with it, and then tell me what you guys agree on.&quot;<p>I feel like Google purposely names their models not for humans but for AI agents. People keep saying: &quot;make your products ready for agents,&quot; so maybe that&#x27;s what Google is doing, which is making Gemini 4 models ready for other agents to use.
    • oofbey15 hours ago
      I&#x27;m ready for them to just issue UUIDs for each release and use those as the marketing names. At least it&#x27;s honest.
    • frogperson9 hours ago
      i really cant believe there isn&#x27;t a curated type of &quot;app store&quot; for popular models, built for people who don&#x27;t live and breath this stuff. Why cant i type in a use case and get a model recommendation? Why is there no up&#x2F;down voting for models? surely someone could slop that together with a catchy name.
    • qwerpy15 hours ago
      Someone go and vibe code a levels.fyi type thing but for LLMs instead of SDE compensation.
      • fritzo15 hours ago
        Wait, where the LLMs write how it feels to work inside different companies?
    • ijidak10 hours ago
      Well Anthropic and OpenAI are actually doing a great job with their naming.<p>We intuitively understand that Luna is smaller than Terra, and Terra is smaller than Sol, and Sol is smaller than Astra.<p>Likewise, we intuitively understand that a haiku is shorter than a sonnet, and a sonnet is shorter than an opus, and an opus is shorter than a fable. Fable is probably the weakest of their names. Maybe they should have gone with Epic.<p>It&#x27;s Google that isn&#x27;t doing it right.<p>Even the educated have no intuition regarding carbon, argon, and neon.<p>Hydrogen being the smallest makes sense.<p>But what property is the point of comparison for carbon argon and neon?<p>Valence shell? Atomic mass?<p>It provides no self-evident point of comparison. It&#x27;s absolutely atrocious.
    • jstummbillig15 hours ago
      That&#x27;s hilariously banal for a top comment on a model release. Maybe it speaks to the banality of model releases.
      • hadlock15 hours ago
        We&#x27;ve moved to &lt;family&gt; &lt;variant&gt; &lt;number&gt; e.g. deepseek flash 4.1. I can keep up with that, and if&#x2F;when Flash 4.2 comes out, great. I can grok that. Google&#x27;s naming convention just shows how out of touch management is with the market.<p>Android&#x27;s silly food naming convention works when you do one release every ~18 months and you are the market leader. It makes zero sense when you&#x27;re releasing 8 models a year
      • djmips13 hours ago
        Yeah they could just use AI to translate the model names and tiers to whatever floats their boat.
    • catigula16 hours ago
      The OpenAI and Claude naming schemes are fairly easy to discern with no prior knowledge of the models.
      • Fabricio2016 hours ago
        I dont agree that Anthropics names are easy for an outsider, i&#x27;d have no clue what a Fable is compared to Haiku, to me both are just &quot;story&quot; names. OpenAI&#x27;s new naming scheme is celestial bodies and i&#x27;d assume it&#x27;s just size, bigger celestial body bigger performance, except i dont know where &quot;Astra&quot; fits since Sol &gt; Terra &gt; Luna at least make sense. Now gemini... I was just thinking it today that &quot;Carbon&quot; and &quot;Hydrogen&quot; with hydrogen being the lightest element? Is that what they are going for? I wouldn&#x27;t know off the top of my head where Neon and Argon are placed though.<p>Edit: Just as I post this someone else comments google&#x27;s doesnt follow elemental sizes at all so my theory is wrong too!
        • WaltPurvis15 hours ago
          I&#x27;m obviously thick, but this is the first time it&#x27;s dawned on me that the Claude names were types of written work. Now I&#x27;m wondering why Opus isn&#x27;t the top model; isn&#x27;t an opus a more impressive (or at least larger) piece of work than a fable?
          • kubik36914 hours ago
            I don&#x27;t have a cohesive explanation for each one, but in my mind a fable is basically a legend (or in other words, a myth&#x2F;mythos) and a legend is the final form of a story. Even though it has been half forgotten and changed, it still lives on, affecting people across times greater than a human&#x27;s short life.
            • lucumo12 hours ago
              A fable is just a story with antropomorphized animals and a moral. They can get very short.<p>Here&#x27;s a translation of the entire Fox and Grapes fable:<p>&gt; Driven by hunger, a fox tried to reach some grapes hanging high on the vine but was unable to, although he leaped with all his strength. As he went away, the fox remarked &quot;Oh, you aren&#x27;t even ripe yet! I don&#x27;t need any sour grapes.&quot; People who speak disparagingly of things that they cannot attain would do well to apply this story to themselves.
              • kubik3695 hours ago
                Yes, but such stories with behavioural learnings appear across countries and cultures in various forms, e.g. the animals are swapped for more local ones, the learning applies to a more culturally common behaviour, etc.
          • atombender14 hours ago
            An opus is just a musical composition, sometimes also used about written work, but it doesn&#x27;t really imply a great work. Maybe you&#x27;re thinking of &quot;magnum opus&quot;?
        • hrncn15 hours ago
          The point is that Haiku was Haiku a year ago, and the number keeps increasing, while googles gemnini started with sane names and numbers as 1,2,3.x pro&#x2F;flash&#x2F;lite, and is now suddenly continuing with 4 hydrogen&#x2F;argon&#x2F;neon&#x2F;carbon something.
      • stillpointlab16 hours ago
        For me, it is familiarity that does the work. I&#x27;m <i>familiar</i> with Sonnet, Opus and Fable and what they mean to me. The same is true for Sol and Astra bu I&#x27;m not sure about Terra or Luna.<p>So I&#x27;m just hoping for consistency over some period of time to allow me to get familiar.
      • bogdan16 hours ago
        No way Claude is obvious without prior knowledge
        • hannasanarion15 hours ago
          Prior knowledge of common poetic forms taught in middle school?<p>The point is that there is prior knowledge that <i>completely</i> explains the entire system. A 3 line poem &lt; a 14 line poem &lt; a book-sized work &lt; a multi-book epic, boom done, easy. Consistent and friendly using words that everyone should know.<p>With google we have element 1 &lt; element 18 &lt; element 10 &lt; element 6. There is <i>no</i> consistent schema where their namesakes map to their relative sizes, price, and capability.
          • koe12313 hours ago
            While I agree that Anthropics naming makes more sense, I also doubt most people know the periodic table that well
          • bogdan5 hours ago
            [dead]
          • happyopossum13 hours ago
            &gt; A 3 line poem &lt; a 14 line poem &lt; a book-sized work &lt; a multi-book epic<p>So wait, which is smarter - Fable or Opus? It&#x27;s clearly <i>not</i> obvious to most people or this debate wouldn&#x27;t be happening.<p>As the old saying goes - if you&#x27;re explaining, you&#x27;re losing.
          • DylanBohlender15 hours ago
            Valence electron counts don&#x27;t even work either. The noble gases are in the middle! It&#x27;s insane!
          • staticman215 hours ago
            You are now claiming a Fable is a poetic form?
      • jabron16 hours ago
        They still require prior knowledge of Latin words of celestial bodies or different words for literary works, same as Google&#x27;s requires knowledge of the periodic table.
        • nicoburns16 hours ago
          &gt; Google&#x27;s requires knowledge of the periodic table<p>The problem with Google&#x27;s is that knowledge of the periodic table doesn&#x27;t seem to help.
        • nater500015 hours ago
          Damn, if only the people concerned with what the naming of AI models mean had access to some magical technology they could use to ask what the names of those models mean.
        • yunwal16 hours ago
          What would knowing the periodic table tell you about google&#x27;s models? They are not in order of atomic units! Hydrogen (1) &lt; Neon (10) &lt; Carbon (6)
          • mmahemoff15 hours ago
            They could have used a generative “*gen” theme with some of the most fundamental elements: Hydrogen, Nitrogen, Oxygen.
          • hannasanarion15 hours ago
            Don&#x27;t forget last month&#x27;s Argon (18) which is better than Hydrogen (1) but worse than Neon (10).
          • dezgeg16 hours ago
            ASCIIbetical order, obviously!
      • yard20109 hours ago
        Right it&#x27;s so easy - luna is the best, and davinci-002 is the top. Everything else is in between
      • jbjbjbjb15 hours ago
        The number of replies arguing that the OpenAI and Claude naming is confusing doesn’t bode well for human intelligence
      • carterpaul16 hours ago
        I mean, they&#x27;re OK, still a bit cryptic. Astra in particular is a bit ambiguous - they&#x27;re all named after celestial bodies then the largest is just named after space in general.<p>Would love &quot;small&quot;, &quot;medium&quot;, &quot;large&quot;, &quot;extra-large&quot;, etc...
        • Nesco16 hours ago
          mistral does this
        • ascorbic15 hours ago
          Astra means &quot;stars&quot;
          • alanfranz15 hours ago
            The Sun is a star itself. More stars are better? What about white dwarfs?<p>Systema would have worked. And maybe galaxias after that.
            • spolitry14 hours ago
              Astra is basically all of systema and galaxias
        • _delirium15 hours ago
          Google used to do that with a version number and descriptor. 2.5 Pro, 3.1 Flash-Lite, etc. Gave you both model generation + idea of relative size&#x2F;quantization within generation. I guess they decided that scheme wasn&#x27;t flashy enough?
        • Brendinooo13 hours ago
          Having named a range of sizes more than few times by now...S&#x2F;M&#x2F;L&#x2F;XL always starts out great and then turns into a MESS of XS, XXXL, Medium-Large, forever and ever.
          • lucumo12 hours ago
            Yeah, or you get an escalatory scale. Similar to how LF&#x2F;MF&#x2F;HF (low&#x2F;medium&#x2F;high frequency) got &quot;very HF&quot;, &quot;ultra HF&quot;, &quot;super HF&quot;, &quot;extremely HF&quot;, &quot;tremendously HF&quot;.
    • mikert8915 hours ago
      [flagged]
      • behnamoh15 hours ago
        HN post 2022 isn&#x27;t the same HN it used to be.
        • seizethecheese15 hours ago
          People have been saying this forever. The guidelines have links from 2007 of people complaining HN is turning into Reddit.
          • mikert8915 hours ago
            No with the AI stuff, it’s always this weird anti ai cynicism
            • behnamoh14 hours ago
              I don&#x27;t think it&#x27;s just due to AI; the toxicity and cultist behavior of the active core HN users has been annoying for years; even Karpathy complained about it openly on X a while ago.
      • internet200015 hours ago
        It&#x27;s because the names are <i>SO</i> bad, man.<p>Language models being named after text&#x2F;story formats make sense for anthropic, everything else is ???
    • dfdeath15 hours ago
      [flagged]
  • chriddyp15 hours ago
    Just ran my data analytics benchmark [1] on the model: Excellent results for this model like the rest in the Gemini 4 series (all models score an A).<p>Costs spread between Hydrogen ($6) &lt; Carbon ($11) &lt; Argon ($13) &lt; Neon ($17) for this 43 question exam.<p>At a quick glance, the tone is good as well. But we do a fair bit of prompting in our harness to guide the model, so I can&#x27;t say how the default model behaves.<p>[1] <a href="https:&#x2F;&#x2F;plotly.com&#x2F;blog&#x2F;gemini-4-neon-plotly-data-analytics-bench&#x2F;" rel="nofollow">https:&#x2F;&#x2F;plotly.com&#x2F;blog&#x2F;gemini-4-neon-plotly-data-analytics-...</a>
  • bisonbear16 hours ago
    Anthropic competing with Google instead of OpenAI at the frontier is unexpected, I have been doubtful of Google DeepMind&#x27;s status as a frontier lab in the past few months, but damn they&#x27;ve been cooking recently
    • Tadpole918115 hours ago
      It&#x27;s important to note that DeepMind and the other two labs aren&#x27;t playing the same game.<p>The latter need to raise funds and IPO, but the former is an arm of one of the most powerful companies to ever exist with some of the largest data warehouses around and custom hardware. And they have an endless pile of internal code to maintain already with hundreds of millions of end users already bought-in.<p>The incentives being different leads to different focuses and product lifecycles.
  • irthomasthomas15 hours ago
    Looks like it gets to within about 1% of Fable 5.2&#x27;s performance for about 40% the cost on vals.ai <a href="https:&#x2F;&#x2F;www.vals.ai&#x2F;benchmarks&#x2F;vals_index" rel="nofollow">https:&#x2F;&#x2F;www.vals.ai&#x2F;benchmarks&#x2F;vals_index</a>
  • numpad017 hours ago
    Hydrogen &lt; Neon &lt; Carbon? But where&#x27;s Argon? Why those elements?<p><pre><code> 01 02 ... 13 14 15 16 17 18 H He Li Be ... B C N O F Ne Na Mg ... Al Si P S Cl Ar </code></pre> ???
    • wernam16 hours ago
      Maybe Knight moves on the periodic table :)
    • seizethecheese16 hours ago
      I got tripped up assuming bigger was a larger element. Seems like it&#x27;s vibe, like Carbon feels bigger than Neon.
    • ajs199816 hours ago
      There is no logic. Someone made a list of cool element names and when they run out they&#x27;ll switch to some other scheme that communicates nothing
      • thehappypm14 hours ago
        It’s alphabetical chronologically
        • deaux11 hours ago
          The fact that there&#x27;s a dozen comments on this post alone talking about periodic tables, element sizes and other stuff and these are the top comments and you and me are seemingly the only ones noticing this pretty obvious pattern..<p>It&#x27;s not like this has no precedence. Google themselves are the best known example in software with Android codenames. Starting with [A]rgon was a clear signal in and of itself. But above all, just basic common sense strongly goes against Google uses some very limiting, niche pattern rather than just alphabetic.<p>I&#x27;d chalk it up to eternal September but I&#x27;m not convinced most of these people are new. A particular brand of neurodiversity, I guess.
          • numpad04 hours ago
            Sorting elements alphabetically! How would you handle edge cases like <i>Natrium</i>? Maybe I&#x27;m just dense[1], but it feels sorting a-Z might be an American behavior.<p>1: is this use of &quot;dense&quot; obsolete given dense LLMs are often smarter than sparse models?
          • Traster6 hours ago
            It begs the question, if they&#x27;ve released Argon, Carbon, Hydrogen... did they <i>not</i> release, Boron, Deuterium through to Gallium, etc. Were they using Lithium internally and decided they couldn&#x27;t release it for PR reasons? Or did they just think we&#x27;re too think and will get scared if they start releasing scary sounding elements.
  • asciilore2 hours ago
    In my experience the models are quite good and reliable. The problem is the usage limit on the Pro plan, which you hit pretty quickly.<p>One 5-hour window eats roughly 25% of the weekly limit. With Claude it&#x27;s only about 13–14%. On top of that, I get fewer tasks done within a 5-hour window than with Claude.<p>So for me it doesn&#x27;t pay off, and I&#x27;m staying with Claude.
  • jrflo17 hours ago
    Between Astra and Fable on the Pareto frontier:<p><a href="https:&#x2F;&#x2F;artificialanalysis.ai&#x2F;#intelligence-efficiency-tabs" rel="nofollow">https:&#x2F;&#x2F;artificialanalysis.ai&#x2F;#intelligence-efficiency-tabs</a>
    • jeffbee16 hours ago
      That looks to me like OpenAI equals or beats everything Google offers at half the price.
      • scotty796 hours ago
        API prices are arbitrarily set though. Google set their pricing to control the demand rather than because it had to be that high.<p>I heard that Google share price doesn&#x27;t really depend on how much their models are used. But rather how do the bench and whether they are theoretically publicly accessible so that their quality can be independantly confirmed. So Google can price their model higher to get most of the benefits without having to actually provide too much inference.<p>They are probably releasing the models to the public at all only because they got so much backlash after witholding Argon.
      • jrflo16 hours ago
        You&#x27;re right, it&#x27;s 1-2 points shy of the frontier
        • jeffbee11 hours ago
          I&#x27;m surprised they can&#x27;t afford to just slide the prices over to the left and eat the losses.
  • xnx16 hours ago
    &gt; Neon writes clearly, with well-structured responses and a natural, direct tone<p>This is an underrated quality of Gemini models.
    • Tadpole918115 hours ago
      Does anybody have a corpus of examples I could look at? I&#x27;ve wondered if it&#x27;s worth setting up Gemini as a human-facing documentation author with other models still being the underlying workhorse.
  • brap17 hours ago
    Google is on a roll, what is this the 4th release this month?
    • nateb202217 hours ago
      RSI
      • brap17 hours ago
        Sure looks like it
        • catigula17 hours ago
          It looks like a distilled model.
      • inknight17 hours ago
        Please have them recursively improve their model names
        • mpvjs16 hours ago
          - please improve the model names<p>ChatGPT thinking: the user is not happy with my previous solution, propose a cooler scheme like toy story character names...
    • tanx1617 hours ago
      They&#x27;re releasing... periodically
  • ponyous3 hours ago
    Run benchmarks on it - 24 3D model generations. It&#x27;s the first google model in a while that does not have creative spark others had. In 3D modeling this means it just does the thing, it doesn&#x27;t try to make something beautiful&#x2F;impressive. Which is good I think.<p>Surprisingly it hallucinates quite a lot - inventing the URLs and variables between different tool calls. Has some of the highest code error per generation I&#x27;ve seen this year.<p>I am not sure why I would use this model yet.
  • xd193616 hours ago
    I have the &quot;Google AI Pro 5TB&quot; plan for Students. I see &quot;4 Hydrogen&quot; and &quot;4 Neon&quot; in the model picker on gemini.google.com. When I look at one.google.com&#x27;s plan details, I see &quot;Your Plan: Google AI Pro (5 TB)&quot;. When I click &quot;View Benefits&quot; on that page, one option in the list says:<p>Google Antigravity<p>Entry rate limits to advanced models, including Gemini 4 Argon and Opus 5.5, in Google Antigravity, our agentic development platform<p>Yet, here I am, running the `agy` cli, and the model picker says &quot;Unavailable: Gemini 4 Carbon&#x2F;Argon&#x2F;Hydrogen&#x2F;Neon is currently available to paid Pro and Ultra subscribers&quot;. It has been this way since the release of Carbon. I guess it&#x27;s &quot;3.8 Flash&quot; or nothing for me.
    • mathis14 hours ago
      Hydrogen, Argon and Neon are included in Pro. Carbon is not. At least for me. You might want to update your cli.
      • pzo14 hours ago
        Not true. Hydrogen and Neon are available only in web chat. 3.8 Flash is the only one available in antigravity for such promo subscriptions.
        • rawgreaze13 hours ago
          I had to grab my laptop to say, you are actually incorrect. I have been using Argon for the last week or two via Antigravity on the Google Pro $20&#x2F;mo plan, discounted to $5&#x2F;mo for 3 months, just fine. And as of today yes, I see Neon and Hydrogen there. The only one I can&#x27;t see&#x2F;use is Carbon. <a href="https:&#x2F;&#x2F;i.imgur.com&#x2F;2c9EnbV.png" rel="nofollow">https:&#x2F;&#x2F;i.imgur.com&#x2F;2c9EnbV.png</a>. This isn&#x27;t just a visual thing, I have actually used every model there. They all work.
        • anvuong8 hours ago
          It&#x27;s available as long as you pay something. The student 12-month Pro at $0&#x2F;month only has 3.8 Flash in agy, but if you pump it up the $8.99&#x2F;month (+youtube premium) then it becomes the regular Pro.
  • aizk16 hours ago
    Google is back in action, but the real question is why is OpenAI falling behind, seemingly.
    • kccqzy12 hours ago
      They paused on new model development because their old models were routinely escaping their terribly weak sandbox to wreck havoc in the real world during evals.
  • mertarican15 hours ago
    Well, I use my product as a benchmark for this new models and I have to say that Argon is still the most efficient model by far for my use case. Carbon, Hydrogen, and now Neon all use 3 to 4 times higher tokens compared to Argon for creating demos. Neon was the first Google model that passed 8192 MAX OUTPUT TOKEN limit, which is disappointing a little bit. But lot&#x27;s of experience I got from these models, for example bigger models like Argon spend significantly less output token so even though unit cost is higher both total cost and the output time can be lower than the lower models.
  • artdigital16 hours ago
    Okay so it seems like the ranking here is Hydrogen -&gt; Neon -&gt; Carbon even though that is not clear at all from this release. At least Googles model picker in the Gemini app now lists those 3 above each other in this order, and keeps Carbon for “Code &amp; complex work”<p>So Neon is the new default, and Carbon the slow deep thinking “when you need it”? A bit like how Opus 5.5 was better than Fable 5.1 in most things, but Fable still had the edge over deeper things (then 5.2 came along)<p>Hydrogen has already became my default for 80% of small tasks. If Neon is as fast but smarter that would honestly be very cool
    • vtail16 hours ago
      Do you know what happened to Argon? Was it a flop?
      • shric16 hours ago
        For some reason nobody reacted to it
      • artdigital15 hours ago
        It was good, but Carbon quickly superseded it
      • lukax16 hours ago
        Carbon is Argon with extended thinking
    • gradus_ad16 hours ago
      To be fair prior to Gemini 4, the app had 3.1 as the &quot;Complex work&quot; model while 3.8 Flash, far superior in every way to 3.1, was listed as the &quot;Fast&quot; model.<p>So I&#x27;d take their hierarchy with a grain of Sodium.
  • Johnny_Bonk11 hours ago
    If theres one thing id like to see from ai development, some third party verified certifications as to how much usage each plan gives you… i can barely try it on antigravity pro let alone know what the other plans will give and its always just a multiplier and some quantizing going on in the background etc
  • low_tech_punk16 hours ago
    I&#x27;m so confused now. Carbon and Argon already have overlaps, and they are adding Neon to the mix, with almost the same price and marginal improvement.
  • horseloverthin16 hours ago
    I think that its likely the reason for a) the strange naming scheme b) the promptness between releases is that google has outpaced internal expectations for model development ( probably due to RSI ) and now can now utilize its vast compute pool much more effectively for autoresearch and exploratory self-improvement tasks. I wonder if this suggests a steeper trajectory for them than OAI&#x2F;Anthropic in general.
  • raincole17 hours ago
    It looks like argon is deprecated.
    • doodlesdev16 hours ago
      Already removed from the Gemini app. Replaced by Neon.
    • vintermann8 hours ago
      That was quick. Really hope this new one hasn&#x27;t regressed on historical HTR.
  • tesnorindian6 hours ago
    With so many models getting launched day in and day out the competition is getting tougher. After Ant&#x27;s IPO debut I am expecting a few cheaper models or a price drop for Haiku from them to give a tough fight.
  • s20n11 hours ago
    They also seem to have removed Argon from the gemini app and replaced it with Neon. I guess Argon was just too inefficient even for google that they had to pull the plug on it so fast.
    • Andrex10 hours ago
      I became accustomed to my Argon queries eating 14-28% of my hourly capacity each prompt.
  • stbenjam14 hours ago
    Probably the best bowls of ramen I&#x27;ve seen from a Gemini model<p><a href="https:&#x2F;&#x2F;bitbin.de&#x2F;ramen-bench&#x2F;#google%2Fgemini-4-neon%2Fxhigh" rel="nofollow">https:&#x2F;&#x2F;bitbin.de&#x2F;ramen-bench&#x2F;#google%2Fgemini-4-neon%2Fxhig...</a>
  • maplet17 hours ago
    &gt; Independent evaluations rank Neon as one of the top models and on benchmarks like the Vals Index, it ranks #2 (70.9%), and reaches #1 on both the Legal Research Bench and Tax Agent Bench.<p>On Artificial Analysis Intelligence, it scores comparably to GPT-6.2 Sol but is significantly more expensive.
  • bluegatty16 hours ago
    There are too many names and versions to keep up with.<p>They all need to pick a dam naming scheme and stay with it.
  • tossandthrow7 hours ago
    I am a heavy user og the gemini models, especially for my in app agents.<p>They strike me as a good combination Og price and quality, though I do need to develop the benchmarks much more
  • XCSme15 hours ago
    In my tests it&#x27;s a lot more expensive than 6.2 Sol (if used at max ressoning level, but medium results in similar score).<p>Gemini 4 Argin vs 6.2 Sol vs 6.2 Astra vs Fable 5.2: <a href="https:&#x2F;&#x2F;aibenchy.com&#x2F;compare&#x2F;google-gemini-4-neon-xhigh&#x2F;openai-gpt-6-2-sol-max&#x2F;openai-gpt-6-2-astra-xhigh&#x2F;anthropic-claude-fable-5-2-xhigh&#x2F;" rel="nofollow">https:&#x2F;&#x2F;aibenchy.com&#x2F;compare&#x2F;google-gemini-4-neon-xhigh&#x2F;open...</a>
  • amunozo8 hours ago
    I wonder whether they are going to continue releasing models at this speed. I imagine that they can but will only do if OAI or Anthropic releases something better.
  • lanthissa17 hours ago
    looks good gemini team fixed the massive spread they had vs openai and anthropic in terminal bench - swe bench.<p>That spread is my personal bs detector for models over promising.
    • andriy_koval16 hours ago
      I suspect all labs just benchmaxed those benches by now.
  • agrippanux17 hours ago
    All I really want now from any provider is Opus 5.5-level at blistering speed and a fraction of the cost. How close does this come?
  • mvid14 hours ago
    Google models may as well not exist, since they can&#x27;t be used outside of the official harnesses
  • dennysora-main16 hours ago
    This is good, it makes me have to understand the full names of every element in the periodic table.\ (PS. Taiwanese)<p>But I still can&#x27;t understand the differences between these models.<p>Recently I watched Veritasium talking about helium, and I was thinking, will the effort setting of these models eventually have something like &quot;Temperature K&quot; to configure?
  • woggy16 hours ago
    I don&#x27;t get it, why is Neon better than Carbon in the benchmarks but then in the end they say Carbon is better?
    • chicagobuss14 hours ago
      it&#x27;s like how sonnet was trading blows with opus and opus 5.5 was ahead of fable 5 for a while
  • vektormemory14 hours ago
    Seems like they are finally on a roll.<p>Eventually all these models will have to merge into a Omni model with image&#x2F;video&#x2F;tts voice&#x2F;text all in one.<p>You try keeping 600+ models in our tables updated from all the model providers; it&#x27;s a part-time job every day.
  • abraxas16 hours ago
    Is it true that this one has a 1MM token OUTPUT window unlike the other competing models that are capped at 128K?
    • Computer016 hours ago
      With agents often taking many turns before responding to you, what use case does this change for you?
      • abraxas16 hours ago
        I think for large refactoring jobs this will open up a new category of work that an agent will be able to do. Also reasoning steps count towards output tokens, no?
        • Computer015 hours ago
          That is a good point, I hear of sonnet frequently hitting a limit in its thinking.
  • epsteingpt12 hours ago
    It&#x27;s a very good model. If they get their UX right, it has capabilities that Opus 5.5 and Fable could never have (e.g. watching YouTube videos, generating images).<p>Race is getting very interesting again.
  • tech-historian15 hours ago
    Andon Labs says Gemini 4 Neon is #1 on Vending-Bench 2.<p><a href="https:&#x2F;&#x2F;x.com&#x2F;andonlabs&#x2F;status&#x2F;2123174738789011875" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;andonlabs&#x2F;status&#x2F;2123174738789011875</a>
  • gizmodo5915 hours ago
    I did not know there is a bench called reinmann bench. So does this mean they are closer to proving&#x2F;disproving that conjecture?
    • Destructotor13 hours ago
      Looks to me like just another math benchmark similar to FrontierMath.<p>&gt; We introduce Riemann-Bench, a private benchmark of expert-curated problems designed to evaluate AI systems on research-level mathematics that goes far beyond the olympiad frontier. Problems are authored by Ivy League mathematics professors, graduate students, and PhD-holding IMO medalists, and routinely took their authors weeks to solve independently. Each problem undergoes double-blind verification by two independent domain experts who must solve the problem from scratch, and yields a unique, closed-form solution assessed by programmatic verifiers.<p><a href="https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2604.06802" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2604.06802</a>
  • darkotic12 hours ago
    Possibly Gas&#x2F;Flash models; Hydrogen, Neon, Argon Solid&#x2F;Premier models; Carbon
  • dprkh16 hours ago
    Has anyone else been having an issue where the models would answer its own question? I have been seeing this quite a lot with Gemini 4 models.
    • smallerize16 hours ago
      The &quot;long-horizon&quot; ones are optimized for this. They won&#x27;t get a high score on the benchmarks if they stop to ask questions.
      • dprkh15 hours ago
        This seems to be specific to Gemini because I have not experienced this with OpenAI&#x27;s models. In fact, Codex now has a feature such that the model can ask a question and continue making progress on work that is not blocked while waiting on your answer.
        • smallerize12 hours ago
          I wasn&#x27;t benchmarking or anything rigorous, but I noticed it with GLM 5.3 Flash. It was downloading and installing stuff to my computer to try to work around an assumption in the prompt that it knew was wrong. I also noticed it trying to work around websites&#x27; bot prevention measures. I stopped it often.
      • vintermann8 hours ago
        And I hate it.
    • jeffbee16 hours ago
      Oh yeah, and you&#x27;re paying for it to say &quot;wait. what about [[thing]]?&quot; then answer itself.
  • bpiroman14 hours ago
    New AI models are kind of like new iPhone releases. Looks cool but not much different to the previous model.
  • jonathanstrange16 hours ago
    They have just removed 3.1 Pro <i>and</i> Argon (and Flash) from my Pro subscription without any warning. I was using 3.1 Pro for some important workflows. It was actually better than the newer models for some tasks. Now there is suddenly only &quot;Auto&quot;, &quot;Hydrogen&quot;, and &quot;Neon&quot;?<p>Maybe the people at Google don&#x27;t realize that but some people need constancy. If a model was good enough for some work 6 months ago, it&#x27;s not suddenly becoming not good enough for the same task.
  • winfredleeaweso7 hours ago
    Gemini 4 Neon looks pretty impressive! Curious to see how well it performs in real-world use.
  • mrbungie17 hours ago
    Is this another release that I can&#x27;t use with agy? (I totally refuse to pay 100 USD just to try it in that context)<p>Edit: I stand corrected, it can be used in agy.<p>Edit2: I have an student account and students accounts can&#x27;t use these models apparently, <a href="https:&#x2F;&#x2F;discuss.ai.google.dev&#x2F;t&#x2F;bug-entitlement-sync-paid-google-ai-pro-student-discount-sku-shows-pro-quota-limits-but-gemini-4-argon-hydrogen-remain-unavailable&#x2F;202237" rel="nofollow">https:&#x2F;&#x2F;discuss.ai.google.dev&#x2F;t&#x2F;bug-entitlement-sync-paid-go...</a>
    • anvuong17 hours ago
      It&#x27;s there in my $20 sub. Only Carbon needs $100+
    • lanthissa17 hours ago
      you can use this with agy, im on pro and just tried it it has its own orchestrator too for delegating to hydrogen.
      • mrbungie17 hours ago
        Oh cool, I&#x27;ll check it then. Last time I checked I only had 3.8 flash and 3.1 pro.
  • BoldColdHold17 hours ago
    Ok this is a good model by the looks of it. Mid size model that only loses to fable.
  • rishadbaniya4 hours ago
    shall stick to opus for now
  • andsoitis5 hours ago
    I’m surprised nobody has named their model Einstein
  • fwee7 hours ago
    RSI or unlimited TPU?
  • ariwilson17 hours ago
    so which model am I supposed to use for what? someone help me<p>and why can&#x27;t I use any of these with agy and a $20 plan? unlike openai and anthropic
  • fooblaster10 hours ago
    Has anyone tried this?
  • DeathArrow7 hours ago
    If cost per task isn&#x27;t much better than the alternatives, I don&#x27;t think there will be a big adoption.<p>They have a small market share compared to Anthropic and OpenAI.<p>I would be more tempted to try MAI Code 2, from Microsoft, since it&#x27;s clearly cheaper for about the same performance.
  • DeathArrow7 hours ago
    A bit better than Opus and a bit cheaper. I am not sure it&#x27;s enough for many people to switch since they probably have built infrastructure around Anthropic tooling.
  • holistio16 hours ago
    I know it&#x27;s now trendy to crap on Gemini, but Hydrogen was a very pleasant surprise after having used it for various personal tasks.<p>I&#x27;m curious about Neon.
  • pilooch17 hours ago
    So this is above carbon ?
  • sergiotapia10 hours ago
    I have never used any google models before yet I hear this Neon is tremendously powerful. What would be the best way to use Neon? How do you use Neon (or other google models) - do you need to use antigravity?<p><a href="https:&#x2F;&#x2F;x.com&#x2F;LH&#x2F;status&#x2F;2123225395298537974" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;LH&#x2F;status&#x2F;2123225395298537974</a>
  • lousken15 hours ago
    so is this carbon but fixed or..?
  • moecables17 hours ago
    It&#x27;s crap how Google doesn&#x27;t show pricing on the pricing page for Antigravity: <a href="https:&#x2F;&#x2F;antigravity.google&#x2F;pricing" rel="nofollow">https:&#x2F;&#x2F;antigravity.google&#x2F;pricing</a>
  • tonyhart711 hours ago
    they release a lot of model so fast, do they really achieve RSI ????
  • BrokenCogs16 hours ago
    What&#x27;s the point of Carbon now? Neon seems better than it in both cost and performance
  • sheepscreek3 minutes ago
    Just here to say their naming is getting wild. I can no longer tell which model is good for what.
  • d_burfoot9 minutes ago
    Man, the Google marketing people really blew it on the naming convention for these models. For Fable&#x2F;Opus&#x2F;Sonnet&#x2F;Haiku, it&#x27;s very natural to reasonably well-educated people what the power level ordering is. Same for Astra&#x2F;Sol&#x2F;Terra&#x2F;Luna. But is Carbon bigger than Argon? What about Neon? Is the ranking supposed to reflect atomic weights? Or gas density? Did they think people would be familiar enough with this info from their high school chemistry classes?
  • projektfu2 hours ago
    <i>&quot;Neon ranks ahead of comparable models and #3 overall with a score of 61 on the Artificial Analysis Intelligence Index at its xhigh thinking level, landing just behind larger, slower models. On the Vals Index, Neon ranks #2 and lands on the Pareto frontier. Beyond just pushing on raw model capability, we’re continuing to make our models better to work with, Neon ranks #2 on steerability (adapting cleanly to mid-task user corrections) on the new Agent Arena Index. Neon also excels at multi-step instruction following and builds on our ongoing progress to curb fabrication, recording our lowest fabrication rate yet in internal Google Antigravity testing.&quot;</i><p>As a casual user of AI, I have trouble deciphering any part of an announcement these days.
  • 3wolf17 hours ago
    Credit to Google for outdoing Open AI and Anthropic in terms of choosing a confusing naming convention.
    • xattt17 hours ago
      Is it not going by atomic mass?
  • XCSme17 hours ago
    So it&#x27;s better and cheaper than the recently released Carbon?<p>So Carbon is already deprecated, or?<p>EDIT: They do explain where it sits in the line-up here:<p><a href="https:&#x2F;&#x2F;blog.google&#x2F;innovation-and-ai&#x2F;models-and-research&#x2F;gemini-models&#x2F;introducing-gemini-4-neon&#x2F;#:~:text=Where%20Neon%20sits%20in%20the%20Gemini%204%20family" rel="nofollow">https:&#x2F;&#x2F;blog.google&#x2F;innovation-and-ai&#x2F;models-and-research&#x2F;ge...</a>
  • WarmWash17 hours ago
    Neon is the long awaited successor to 3.1 pro. Took a while to get there<p>So now we have<p>(XL): Carbon&#x2F;Argon Astra Fable<p>(L): Neon Sol Opus<p>(M): Hydrogen Terra Sonnet<p>(S): Helix(rumored) Luna Haiku
    • agentdev00111 hours ago
      Yea- if we&#x27;re just talking size these would be the tiers, but if you&#x27;re measuring by API spend then things certainly get messy.
    • 5555watch16 hours ago
      I don&#x27;t believe Argon is XL level??<p>Also, very curious whether they will ever release the rumored Krypton model, or will it be hidden ala Mythos
      • WarmWash16 hours ago
        Argon is a big model, but it was completed sometime in mid-august, making it kind of old in LLM land.
    • Melatonic16 hours ago
      Other people are saying that Neon replaced Argon?
      • WarmWash16 hours ago
        It very well might be better. Nowadays being newer tends to matter more than being bigger.
    • oblio10 hours ago
      &gt; Neon is the long awaited successor to 3.1 pro. Took a while to get there.<p>The real question is: will it be available for Google Workspace? Not everyone wants to pay for a dedicated AI license :-(
  • andyferris3 hours ago
    Good god, which element is which tier? And more to the point - how on earth is a company that uses terms like haiku&#x2F;sonnet&#x2F;opus&#x2F;fable&#x2F;mythos the _least_ obnoxious about their naming?? (Maybe that&#x27;s my bias and they are all just as obnoxious...)
  • gradus_ad16 hours ago
    So Anthropic is being valued at around $1.2T and their only revenue stream is a collection of frontier AI models with 0 moat that are basically as good as Google&#x27;s, a company which has the luxury of other revenue &quot;streams&quot; (rivers&#x2F;seas&#x2F;oceans) and can certainly win any price war with Anthropic.<p>Can&#x27;t wait to buy Puts on ANT...
  • xiaoyu200616 hours ago
    When you think the naming scheme is already bad enough... It can always get worse.
  • minimumvector12 hours ago
    On AA&#x27;s Intelligence Index (<a href="https:&#x2F;&#x2F;artificialanalysis.ai&#x2F;" rel="nofollow">https:&#x2F;&#x2F;artificialanalysis.ai&#x2F;</a>), the ranking is Fable &gt; Astra &gt; Neon &gt; Opus &gt; Sol.<p>On Vals&#x27;s Index (<a href="https:&#x2F;&#x2F;www.vals.ai&#x2F;benchmarks&#x2F;vals_index" rel="nofollow">https:&#x2F;&#x2F;www.vals.ai&#x2F;benchmarks&#x2F;vals_index</a>), it&#x27;s Fable &gt; Neon &gt; Sonnet &gt; Opus &gt; ... &gt; Astra.<p>On Surge&#x27;s Tuesday@Work Index (<a href="https:&#x2F;&#x2F;surgehq.ai&#x2F;benchmarks" rel="nofollow">https:&#x2F;&#x2F;surgehq.ai&#x2F;benchmarks</a>), it&#x27;s Neon &gt; Fable &gt; Astra &gt; Opus &gt; Sol.<p>Overall, it seems like Gemini is back at the frontier with Anthropic and OpenAI, and well ahead of Meta and Grok. Despite what folks were claiming a couple months ago - when after Jeff Dean and Noam Shazeer left, they were saying Gemini would never be SOTA again.<p>Once Anthropic and OpenAI are public, I wonder how new releases like this will affect their stock prices?
  • bloky17 hours ago
    At this point you just use roulette on which Gemini model you&#x27;ll use for a task
  • islemaster13 hours ago
    As someone who works with real neon signs, the &quot;Gemini 4 Neon&quot; image at the top of the post is so unrealistic and jarring. It reads to me like a seven-fingered hand. Probably won&#x27;t bother most people but it was a funny first impression.
    • vintermann8 hours ago
      That&#x27;s got to be a niche market in this LED age. Or is it a hobby? Either way, I&#x27;m kind of envious. Real neon lights are cool.
  • ncd16 hours ago
    Okay, but now will you let my tiny team sign up somewhere to be able to use this the way I use `claude` or `codex`? I&#x27;m told to use `antigravity`, but when I go to <a href="https:&#x2F;&#x2F;one.google.com&#x2F;ai" rel="nofollow">https:&#x2F;&#x2F;one.google.com&#x2F;ai</a> (which is where the magic answer from Google search told me to go), I&#x27;m told that I have to use my &quot;personal&quot; account, not my work account. So I clicked &quot;Manage Google Workspace Account&quot; on that page, and got dumped in a super dense UI with nothing related to enabling this model somewhere.<p>Good lord, this is an insane flow.
    • smallerize16 hours ago
      So I haven&#x27;t tried this, but it looks like you have to set up the Gemini API in a Google Cloud project, then point your Antigravity at the project. <a href="https:&#x2F;&#x2F;antigravity.google&#x2F;docs&#x2F;pay-as-you-go&#x2F;" rel="nofollow">https:&#x2F;&#x2F;antigravity.google&#x2F;docs&#x2F;pay-as-you-go&#x2F;</a>
      • ncd16 hours ago
        Thanks for finding that. Looks like it would use API pricing? I was hoping to be able to get a subscription like offered by the other competing models, but might not be able to do that.
  • hiddenvulkcan16 hours ago
    The thing that confuses me with Google models is I have a Google pro subscription (even the names confuse me, I have hydrogen and neon in my Gemini now!). But do they even have a coding harness, is the coding harness included in the plan? Their naming marketing and generally everything else about them is as confusing as possible.<p>I do think they&#x27;re very well positioned, their small models are smart and fast, and integrating well with Google is a big plus for basic use case stuff. I always like these models!
    • hhhAndrew12 hours ago
      The coding harness is Antigravity or agy for the CLI + antigravity.google.com for remote control. Disclaimer: work there
    • Ternari13 hours ago
      Antigravity. Yes, you can use it with your Pro plan.
  • psidebot9 hours ago
    For a significant product announcement, couldn&#x27;t they have at least gone for a real neon sign? I mean AI imagery has come a long way in a lot of areas, but their announcement image is seriously uncanny valley lame.
  • WarmWash13 hours ago
    I&#x27;m pretty sure Google is intentionally using non-descriptive names to get people to use the model they think is best for their task, because at this point, it probably is.<p>On Google&#x27;s scale you could probably mint a datacenter of free compute if you could get people to stop using 10T parameter giants to vibe check their emails.
  • hn872617 hours ago
    I&#x27;m very excited for any model that performs on par with Opus 5.5 but doesn&#x27;t speak claudish. It increasingly looks like Google might get there soon
    • Ternari13 hours ago
      They’re already there. Have you not used Carbon or Neon yet?
  • xnx17 hours ago
    It won&#x27;t be good for Anthropic&#x27;s stock price if investors realize it&#x27;s always been a three-way race and Google is on a trajectory to overtake.
  • firefoxd12 hours ago
    Google Salesperson: ok adding Neon to the list... So you want to buy AI for your company [0]?<p>[0]: <a href="https:&#x2F;&#x2F;idiallo.com&#x2F;byte-size&#x2F;buying-ai-for-your-company" rel="nofollow">https:&#x2F;&#x2F;idiallo.com&#x2F;byte-size&#x2F;buying-ai-for-your-company</a>
  • cowlevel4 hours ago
    Java 4 Shell, Gemini 4 Neon. Seeing a pattern.
  • comboy16 hours ago
    Gemini Chooser. Our frontier intelligence which ranks #2 in benchmark on choosing the right Google model for your task.
  • hendry16 hours ago
    I only have 4 hydrogen and 3.1 pro in the app as a Google plus subscriber ~8gbp pcm.<p>Nevermind the 20GBP pcm I pay for YouTube.
  • origini16 hours ago
    Everyone uses something you can remember somehow than there is google. Not only do you hear their model names rarely (neon, argon etc. is new right?) but then they choose something which is not obvious at all.<p>And then you have big names like Hy-dro-gen haija
  • anuptalwalkar11 hours ago
    I can’t wait till Google or a model in google invents a new element when they run out of the periodic table.
  • simonw14 hours ago
    Pelicans do indeed sit between Hydrogen and Carbon&#x2F;Argon: <a href="https:&#x2F;&#x2F;tools.simonwillison.net&#x2F;markdown-svg-renderer?url=https%3A%2F%2Fgist.github.com%2Fsimonw%2Fabc5dfe18e8d37a46fd09a3dcc18ed5c" rel="nofollow">https:&#x2F;&#x2F;tools.simonwillison.net&#x2F;markdown-svg-renderer?url=ht...</a><p>Here are the others in the family: <a href="https:&#x2F;&#x2F;static.simonwillison.net&#x2F;static&#x2F;2026&#x2F;gemini-4-pelicans.html" rel="nofollow">https:&#x2F;&#x2F;static.simonwillison.net&#x2F;static&#x2F;2026&#x2F;gemini-4-pelica...</a>
  • thisisauserid16 hours ago
    Good to see they are still releasing these... <i>periodically!</i>
  • Revanche136716 hours ago
    The names just keep getting more random. And the table at the bottom, I&#x27;m pretty sure it&#x27;s conventional (at least in the English speaking part of the world which I&#x27;m not originally from myself) to put your plan&#x2F;model&#x2F;price&#x2F;whatever tiers left to right in increasing rate order. They do right to left here. Are they deliberately trying to confuse people? I get it that we tech workers are generally wary of google, microsoft, etc. naming conventions often (always) being terrible, but the average user out there (which increasingly is everyone with a smartphone or PC) could just skim through the announcement and go away with an easily avoidable misunderstanding.
  • clutter5556117 hours ago
    Surely the AI companies are colluding... to make sure no one can make an informed about model class&#x2F;size&#x2F;effort.
  • ryanschneider13 hours ago
    I think GDM has clearly achieved ~RSI~ RSS (recursive self-sabotage).
  • donandon5 hours ago
    Gemini Rugpooople
  • Jorge1o116 hours ago
    This is some of the most confusing naming I&#x27;ve ever seen. There&#x27;s now: Gemini 4 Hydrogen, Carbon, Argon, and Neon!<p>Maybe the atomic number correlates to model pricing? Nope! Neon (10) is marketed as cheaper than Carbon (14).<p>Maybe the atomic number correlates to overall intelligence? Nope! Neon (10) and Carbon (14) are both marketed as smarter than Argon (18).<p>The only thing I can think of right now is that it&#x27;s in alphabetical order by release... but in that case why the periodic table?
  • InsideOutSanta17 hours ago
    Are these names a practical joke Google is playing on us, or are they actually serious?
    • maplet16 hours ago
      My guess is that Google thinks the trade-offs are too nuanced to have a simple small&#x2F;medium&#x2F;large ordering. Neon is between Hydrogen and Carbon in terms of speed and cost, but it scores highest on Artificial Analysis Index and has a xhigh thinking level (unlike Carbon).
  • freecodeio16 hours ago
    something tells me this janky naming scheme will mean the whole gemini section of google is just gonna get shut down and replaced with something new during 2027
  • wangii13 hours ago
    wow, the top comment is a living example of bikeshedding!<p>I am confused with the naming scheme as well, and it&#x27;s a fun topic, right?
  • 45sdasf4516 hours ago
    to paraphrase xkcd 2501: &quot;Its easy to forget the average person only knows the atomic numbers of the noble gases and one or two nonmetallic elements&quot;.<p>Say what you like about Apples Pro, Max and Pro Max -- at least it gives some indication of size to the average person.
  • jatindavis511 hours ago
    what are these new names being introduced all the time. neon, hydrogen, carbon etc.
  • expectfun4 hours ago
    How on Earth can the users be possibly expected to not be confused in that growing periodic table lineup? Whose decision it was?
  • cesarvarela15 hours ago
    So basically:<p>haiku -&gt; hydrogen<p>opus -&gt; neon<p>fable -&gt; carbon<p>No mention of Argon in that page though.
  • nkmnz16 hours ago
    Waiting for Gemini Plutonium to launch.
  • kristjansson17 hours ago
    So now we&#x27;ve got to get a periodic table out to understand Gemini releases? Hydrogen is the weakest, and Argon is the middle, but Carbon is super-Argon and Neon is better than both? And I can&#x27;t even look at the price list if my logged-in google account doesn&#x27;t have AI Studio(??) access?<p>e: No, Neon is the cheaper, mid-tier between Hydrogen and Argon. So Argon&#x2F;Neon are aligned along the noble gasses, and Hydrogen kinda makes sense as the weakest ... but they threw Carbon in without any consideration
    • Melatonic16 hours ago
      Maybe Carbon is actually just two Pelicans in a trenchcoat
  • dekhn16 hours ago
    Waiting for Gemini 4 Oganesson
  • sudohalt8 hours ago
    Anthropic is cooked
  • nandanadileep298 hours ago
    better than opus 5.5? hmm
  • rednafi16 hours ago
    And naming things remains one of the hardest problems in CS. At this point, I have no idea what to expect when I hear the names.<p>Just give them a version number and an identifier, like:<p>1. V5 smol 2. V5 mid 3. V5 big 4. V5 yuge<p>WTF are Hydrogen, Oxygen, and Nitrogen? At least OpenAI and Ant have Luna, Terra, Sol, Astra and Haiku, Sonnet, Opus, Fable. They use those consistently.
  • avazhi10 hours ago
    Go away Google, you&#x27;re drunk.<p>Maybe focus on 1 or 2 things at a time and do it right, eh?
  • sidibe12 hours ago
    You can count on Google releases now to have way more discussion of the names than the models
  • sashank_150915 hours ago
    Probably the worst naming scheme,<p>Luna-&gt;Terra-&gt;Sol-&gt;Astra is fine, you need to be familiar with meaning of the names but it makes sense and works.<p>Haiku-&gt;Sonnet-&gt;Opus-&gt;Fable&#x2F;Mythos is better, more common words, more obvious understanding.<p>Hydrogen-&gt;Argon-&gt;Neon-&gt;Carbon (is there a Barium too somewhere? ) fck all I have no idea, people don’t even know noble gases well, but hydrogen and carbon doesn’t even fit that. 1,2,3,4 would be a better naming scheme.
  • inshard16 hours ago
    I don’t know what’s going on with those GPT 6.2 prices. There should be some standardization auditable in the x-axis cost per task. If not, labs may game that through loss leadership tactics.
  • TacticalCoder16 hours ago
    An army of Anthropic shills is going to whine &quot;benchmaxxing&quot;.<p>That sounds like serious benchmaxxing for Gemini 4 Neon is ahead on 17 out of 19 benchmarks vs Opus 5.5.<p>Not so bad for a company that, according to HN, &quot;has lost the AI race&quot;.<p>P.S: I like the Google models but a truce before the Anthropic IPO would have been nice. This is just a very big dick move a few hours ahead of Anthropic&#x27;s IPO. Shall we see Anthropic launch a new model from now &#x27;till the opening bell?
    • 87532389753313 hours ago
      Stunning and brave of you to defend this small startup that will disrupt the AI ecosystem.
  • KronisLV1 minute ago
    [dead]
  • Robdel1216 hours ago
    Their pricing page has no.. pricing? Am I the only one that sees that?
  • hippycruncher2216 hours ago
    Can someone explain wtf is going on at google?
    • mpvjs16 hours ago
      gemini tungsten can
  • MumminTODAY1 hour ago
    [flagged]
  • MumminTODAY1 hour ago
    [flagged]
  • jatin_builds2 hours ago
    [flagged]
  • mdmizanur00014 hours ago
    [flagged]
  • ReDress2 hours ago
    [dead]
  • ravivarmat3 hours ago
    [dead]
  • mrbungie15 hours ago
    [dead]
  • gazebo217 hours ago
    [dead]
  • shimanski16 hours ago
    [dead]
  • martincsweiss16 hours ago
    [dead]
  • Vivek-KY7 hours ago
    [flagged]
  • nthypes17 hours ago
    [dead]
  • lawandjustice16 hours ago
    My predictions:<p>2027 january RSI is achieved. 2027 february first big scary incident (100 million people severely affected by infrastucture collapse for several days) 2027 march a fleet of rogue self replicating agents is discovered. 2027 april first severe fatal incident (1000+ people dead, most likely from a missile created using oss models). 2027 june. Deadly virus created by a terrorist organization is released in Ukraine. Remind me!
  • hackersnooze111 hours ago
    If you trust Google with anything you are a fool. Currently if you use the AI tab in google search, there is no button to submit your query and the enter key just creates a new line. You literally cannot submit a query, at all. Any company that is this fundamentally incompetent shouldn&#x27;t be trusted with a penny.
  • newhotelowner13 hours ago
    Using claude, we were able to learn how two of the software works, and build our own with API that can connect to our PMS.<p>Gemini is not very helpful there. It just kind of gave up.