17 comments

  • bachittle2 hours ago
    So Opus 4.7 is measurably worse at long-context retrieval compared to Opus 4.6. Opus 4.6 scores 91.9% and Opus 4.7 scores 59.2%. At least they're transparent about the model degradation. They traded long-context retrieval for better software engineering and math scores.
    • film421 hour ago
      To be honest, I think it's just a more honest score of what Opus 4.6 actually was. Once contexts get sufficiently large, Opus develops pretty bad short term memory loss.
    • freedomben2 hours ago
      Agreed, I appreciate the transparency (and Anthropic isn't normally very transparent). It's also great to know because I will change how I approach long contexts knowing it struggles more with them.
      • RobinL2 hours ago
        Could this be because they've found the 1m context uneconomical (ie costs too much to serve, or burns through users quota too quickly causing complaints), and so they're no longer targeting it as a goal
        • Someone12341 hour ago
          Opus 4.7 is also worse at 256K context. Go look at page 195 and page 196. It is across the board regression, not just 1M context.
    • jzig2 hours ago
      At what point along the 1M window does context become "long" enough that this degradation occurs?
      • daemonologist1 hour ago
        The benchmark GP mentioned is measuring at 128k-256k context (there&#x27;s another at 524k-1024k, where 4.6 scored 78.3% and 4.7 scored 32.2%).<p>The longer the context the worse the performance; there isn&#x27;t really a qualitative step change in capability (if there is imo it happens at like 8k-16k tokens, much sooner than is relevant for multi-turn coding tasks - see e.g. this old benchmark <a href="https:&#x2F;&#x2F;github.com&#x2F;adobe-research&#x2F;NoLiMa" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;adobe-research&#x2F;NoLiMa</a> ).
    • teaearlgraycold50 minutes ago
      A year ago it felt like SoTA model developers were not improving so much as moving the dirt around. Maybe we’re in another such rut.
  • kube-system1 hour ago
    &gt; Chemical and biological weapons threat model 2 (CB-2): Novel chemical&#x2F;biological weapons production capabilities. A model has CB-2 capabilities if it has the ability to significantly help threat actors (for example, moderately resourced expert-backed teams) create&#x2F;obtain and deploy chemical and&#x2F;or biological weapons with potential for catastrophic damages far beyond those of past catastrophes such as COVID-19.<p>That&#x27;s an interesting choice of benchmark for measuring the risk of &quot;Chemical and biological weapons&quot;
    • Aboutplants59 minutes ago
      Gotta prime those Government fears!
  • koehr3 hours ago
    This reads more like an advertisement for Mythos, on the first glance
    • Uehreka54 minutes ago
      I never understand these critiques. If something is useful and you’re selling it, does that mean any technical document describing its usefulness becomes marketing?<p>I guess maybe, but then do those documents lose value as technical documents? Not necessarily at all, so I don’t see the point. How are you supposed to describe a useful technical thing to users?
      • parsimo20108 minutes ago
        This is supposedly the Opus 4.7 model card. It&#x27;s okay for it to be marketing for Opus 4.7 and describe what it can do, and even okay for it to talk about what it does better than the last generation. GP was saying it sounds like marketing for Mythos (a different and unreleased model). I don&#x27;t want the Opus 4.7 model card to be advertising for something else.<p>For context, the word &quot;Mythos&quot; appears 331 times in a 221 page document. &quot;Opus 4.6&quot; appears 240 times, so a reference to a model that nobody has really used happens more often than the reference to the last generation model.
    • ModernMech1 hour ago
      That&#x27;s why I don&#x27;t like these &quot;model cards&quot; being presented as if they are some sort of technical document -- they&#x27;re marketing materials.
  • Symmetry2 hours ago
    &gt; The technical error that caused accidental chain-of-thought supervision in some prior models (including Mythos Preview) was also present during the training of Claude Opus 4.7, affecting 7.8% of episodes.<p>&gt;_&gt;
  • aliljet3 hours ago
    Have they effectively communicated what a 20x or 10x Claude subscription actually means? And with Claude 4.7 increasing usage by 1.35x does that mean a 20x plan is now really a 13x plan (no token increase on the subscription) or a 27x plan (more tokens given to compensate for more computer cost) relative to Claude Opus 4.6?
    • computomatic3 hours ago
      They have communicated it as 5x is 5 x Pro, and 20x is 20 x Pro (I haven’t looked lately so not sure if that’s changed).<p>They have also repeatedly communicated that the base unit (Pro allotment) is subject to change and does change often.<p>As far as I can tell, that implies there is no guarantee that those subscriptions get some specific number of tokens per unit of time. It’s not a claim they make.
    • DonsDiscountGas2 hours ago
      Definitely 13x, at least for now
    • ModernMech1 hour ago
      Feels like buying toilet paper.
  • STRiDEX3 hours ago
    Dumb question but why are chemical weapons always addressed as a risk with llms? Is the idea that they contain how to make chemical weapons or that they would guide someone on how?<p>Would there not already be websites that contain that information? How is an llm different, i guess, from some sort of anarchist cookbook thing.
    • somesortofthing22 minutes ago
      They contain broad overviews(throw some disease-causing bacteria in a sort of rainbow arrangement of increasingly more effective antibiotics, you&#x27;ll usually get something that&#x27;s at least very deadly even if it doesn&#x27;t have pandemic potential) but executing in a real lab takes a ton of trial and error to figure out the details. The issue is that the details ~all exist somewhere in the training dataset already, discovered and documented over the course of unrelated, benign biology research. Ability to quickly and accurately search over that corpus translates to large speedups in the physical development process.
    • Philpax3 hours ago
      Both. There&#x27;s the risk of them instructing a user on how to produce a known formulation (the Anarchist Cookbook solution, as you say), which is irritating but not that problematic.<p>The bigger issue is that they are potentially capable of producing novel formulations capable of producing harm, and guiding someone through this process. That is, consider a world in which someone with malicious desires has access to a model as capable at chemistry &#x2F; biology as Mythos is at offensive cybersecurity abilities.<p>This is obviously limited by the fact that the models don&#x27;t operate in the physical world, but there&#x27;s plenty of written material out there.
      • rogerrogerr2 hours ago
        The world has been blessed by two connected things:<p>1. Smart people have economic opportunities that align them away from being evil<p>2. People who are evil tend not to be smart.<p>We&#x27;re breaking both of these assumptions.
        • chrisweekly2 hours ago
          &quot;Smart people have economic opportunities that align them away from being evil&quot;<p>For some definition of evil, some of the time, ok. But as economic opportunities compound (looking at the behavior of the ultra-rich), it seems there&#x27;s at least strong correlation in the other direction, if not full-on &quot;root of all evil&quot; causation.
          • rogerrogerr2 hours ago
            Sure, but that’s not “slaughter a stadium of people with drones” evil or “poison the water supply” evil or “take out unprotected electrical substations” evil.<p>So much infrastructure is very soft because the evil people aren’t smart enough to conceive of or conduct an attack.
            • fwip42 minutes ago
              I think you might find that, if you reconsider who the &#x27;evil&#x27; people are, you might find that we&#x27;re already doing that sort of thing.
        • Der_Einzige2 hours ago
          Good. This is how we will force the world to reckon with the isolated, the disgruntled, and &quot;lone wolf&quot; terrorist. Real &quot;sigma males&quot; actually exist, and when they decide &quot;society has to pay&quot; we are all worse off for it. If Ted Kaczynski (quintessential example of a real actual sigma) had been in his prime operating right now, he&#x27;d have mail-bombed NeurIPS and ICLR already. I&#x27;m not cool with being in crowds of AI professionals right now for physical security reasons given the extreme anti-AI sentiment that exists from nearly everyone outside of the valley: <a href="https:&#x2F;&#x2F;jonready.com&#x2F;blog&#x2F;posts&#x2F;everyone-in-seattle-hates-ai.html" rel="nofollow">https:&#x2F;&#x2F;jonready.com&#x2F;blog&#x2F;posts&#x2F;everyone-in-seattle-hates-ai...</a>
        • malcolmgreaves1 hour ago
          That’s not quite true. Take a look at all the billionaires destroying society. Being evil is the surest way to get to get rich. In fact it’s the only way to amass that level of capital: there’s no ethical billionaire.
          • mikek1 hour ago
            This feels like a wild overgeneralization. People can become rich without resorting to evil methods, especially now with global markets and software. Case in point: Minecraft was wildly successful, and now Notch is a billionaire.
            • hxugufjfjf55 minutes ago
              Eeeeh not the best example maybe?
              • orneryostrich35 minutes ago
                Pre-wealth, Notch was friendly, kind, and downright jolly! Even as he started to accumulate wealth, he was donating huge sums of money to various indie games. Whenever a Humble Bundle dropped he would top the leaderboard for the amount he paid for the games. Things took a major turn for the worse after the acquisition and after he left Mojang. That&#x27;s when he ran out of purpose and turned to drugs and conspiracy theories.
    • Aboutplants58 minutes ago
      It’s marketing, Fear is one of the most effective marketing tools. That and purpose of government attention
    • dcre2 hours ago
      LLMs can tell you exactly how to acquire the materials and manufacture the materials. They might even come up with novel formulations that rely on substances that are easier to get. There might be information about this stuff online but LLMs are much better than random idiots at adapting that information to their actual situation.<p>On top of LLMs reducing the cost&#x2F;difficulty, the other reason biological and chemical weapons are such a worry is their asymmetric character — they are much much easier and cheaper to produce and deploy than they are to defend against.
    • rgbrenner2 hours ago
      In the same way that all coding docs are available publicly
    • CodingJeebus3 hours ago
      WAG but I wonder if a hijacked LLM could also assist with figuring out how to obtain required materials, not just provide the recipe.
  • msla48 minutes ago
    PDF, because it isn&#x27;t marked.
  • deflator1 hour ago
    Model Welfare? Are they serious about this? Or is it just more hype? I really don&#x27;t trust anything this company says anymore. &quot;We have a model that is too dangerous to release&quot; is like me saying that I have a billion dollars in gold that nobody is allowed to see but I expect to be able to borrow against it.
  • 100ms3 hours ago
    <p><pre><code> $ pbpaste | wc -w 62508 $ pbpaste | grep -oi mythos|wc -w 331 $ pbpaste | grep -oi opus|wc -w 809</code></pre>
  • bicepjai3 hours ago
    This card is a 272 page report. So now we are redefining names :)
    • albert_e3 hours ago
      Does the model card fit in the model&#x27;s context :)
      • anonyfox1 hour ago
        well it will saturate your 5h limit window at least
  • joeumn3 hours ago
    I&#x27;m actually surprised at how it performed compared to 4.6 and also compared to mythos. Will be fun to use.
  • il-b2 hours ago
    Ironically, the website is down
  • jmward013 hours ago
    Haiku not getting an update is becoming telling. I suspect we are reaching a point where the low end models are cannibalizing high end and that isn&#x27;t going to stop. How will these companies make money in a few years when even the smallest models are amazing?
    • blixt3 hours ago
      Isn&#x27;t it pretty common for the smaller models to release a little while after the bigger ones, for all the big model providers?
      • jmward013 hours ago
        The last update for Haiku was in October, or in startup land, 10 years ago.
    • mvkel3 hours ago
      It seems to be a rule that older models are more expensive than newer ones. The low end models have higher $CPT and worse output. I wonder if the move is to just have one model and quantize if you hit compute constraints
      • deaux2 hours ago
        &gt; It seems to be a rule that older models are more expensive than newer ones.<p>It isn&#x27;t. Gemini has gotten more expensive with each release. Anthropic has stayed pretty similar over time, no? When is the last time OpenAI dropped API prices? OpenAI started very high because they were the first, so there was a ton of low hanging fruit and there was much room to drop.
    • dkhenry3 hours ago
      The Gemma models are at this point. A 31B model that can fit on a consumer card is as good as Sonnet 4.5. I haven&#x27;t put it through as much on the coding front or tool calling as I have the Claude or GPT models, but for text processing it is on par with the frontier models.
      • make33 hours ago
        absolutely not on par you&#x27;re smoking
        • dkhenry2 hours ago
          You make a compelling argument, but thankfully I have data to back up my anecdotal experience<p>This comparison shows them neck and neck <a href="https:&#x2F;&#x2F;benchlm.ai&#x2F;compare&#x2F;claude-sonnet-4-5-vs-gemma-4-31b" rel="nofollow">https:&#x2F;&#x2F;benchlm.ai&#x2F;compare&#x2F;claude-sonnet-4-5-vs-gemma-4-31b</a><p>As Does this one <a href="https:&#x2F;&#x2F;llm-stats.com&#x2F;models&#x2F;compare&#x2F;claude-sonnet-4-6-vs-gemma-4-31b-it" rel="nofollow">https:&#x2F;&#x2F;llm-stats.com&#x2F;models&#x2F;compare&#x2F;claude-sonnet-4-6-vs-ge...</a><p>And the pelican benchmark even shows them pretty close <a href="https:&#x2F;&#x2F;simonwillison.net&#x2F;2026&#x2F;Apr&#x2F;2&#x2F;gemma-4&#x2F;" rel="nofollow">https:&#x2F;&#x2F;simonwillison.net&#x2F;2026&#x2F;Apr&#x2F;2&#x2F;gemma-4&#x2F;</a> <a href="https:&#x2F;&#x2F;simonwillison.net&#x2F;2025&#x2F;Sep&#x2F;29&#x2F;claude-sonnet-4-5&#x2F;" rel="nofollow">https:&#x2F;&#x2F;simonwillison.net&#x2F;2025&#x2F;Sep&#x2F;29&#x2F;claude-sonnet-4-5&#x2F;</a><p>Also this isn&#x27;t a fringe statement, you can see most people who have done an evaluation agree with me
          • jmward011 hour ago
            I think one area I find hard to get around is context length. Everything self hosted is so limited on length that it is marginal to use. Additionally I think that the tools (like claude code) are clearly in the training mix for Anthropic&#x27;s models so they seem to get a boost over other models pushed into that environment. That being said, open source and local inference is -really- good and only going to get better. There is no doubt that the current frontier biz model is not sustainable.
        • lostmsu3 hours ago
          Just to be clear, did you notice the parent said 4.5?
          • cmorgan312 hours ago
            They are also on par in a lot of classification tasks. I did have to actually use gemma4 and fine tune it a bit but that is part of the value add.
  • NickNaraghi2 hours ago
    232 pages is bullshit. <i>Longer</i> than the Mythos system card? What are you hiding.
  • nothinkjustai2 hours ago
    How much do you want to bet this is Mythos, and Anthropic released it as Opus to avoid embarrassment after all the hype they whipped up…
  • pukaworks1 hour ago
    [dead]