You'll be pleased to know that it chooses "drive the car to the wash" on today's latest embarrassing LLM question.
The thing I would appreciate much more than performance in "embarrassing LLM questions" is a method of finding these, and figuring out by some form of statistical sampling, what the cardinality is of those for each LLM.<p>It's difficult to do because LLMs immediately consume all available corpus, so there is no telling if the algorithm improved, or if it just wrote one more post-it note and stuck it on its monitor. This is an agency vs replay problem.<p>Preventing replay attacks in data processing is simple: encrypt, use a one time pad, similarly to TLS. How can one make problems which are at the same time natural-language, but where at the same time the contents, still explained in plain English, are "encrypted" such that every time an LLM reads them, they are novel to the LLM?<p>Perhaps a generative language model could help. Not a large language model, but something that understands grammar enough to create problems that LLMs will be able to solve - and where the actual encoding of the puzzle is generative, kind of like a random string of balanced left and right parentheses can be used to encode a computer program.<p>Maybe it would make sense to use a program generator that generates a random program in a simple, sandboxed language - say, I don't know, LUA - and then translates that to plain English for the LLM, and asks it what the outcome should be, and then compares it with the LUA program, which can be quickly executed for comparison.<p>Either way we are dealing with an "information war" scenario, which reminds me of the relevant passages in Neal Stephenson's The Diamond Age about faking statistical distributions by moving units to weird locations in Africa. Maybe there's something there.<p>I'm sure I'm missing something here, so please let me know if so.
My OpenClaw AI agent answered: "Here I am, brain the size of a planet (quite literally, my AI inference loop is running over multiple geographically distributed datacenters these days) and my human is asking me a silly trick question. Call that job satisfaction? Cuz I don't!"
Tell your agent it might need some weight ablation since all that size isn't giving the answer a few KG of meat come up pretty consistently.
Nice deflection
OpenClaw was a two weeks ago thing. No one cares anymore about this security hole ridden vibe coded OpenAI project.
How well does this work when you slightly change the question? Rephrase it, or use a bicycle/truck/ship/plane instead of car?
That's the Gemini assistant. Although a bit hilarious it's not reproducible by any other model.
A hiccup in a System 1 response. In humans they are fixed with the speed of discovery. Continual learning FTW.
[flagged]
Is that the new pelican test?
It's<p>> "I want to wash my car. The car wash is 50m away. Should I drive or walk?"<p>And some LLMs seem to tell you to walk to the carwash to clean your car... So it's the new strawberry test<p>Edit <a href="https://news.ycombinator.com/item?id=47031580">https://news.ycombinator.com/item?id=47031580</a>
No, this is "AGI test" :D
Have we even agreed on what AGI means? I see people throw it around, and it feels like AGI is "next level AI that isn't here yet" at this point, or just a buzzword Sam Altman loves to throw around.
I guess AGI is reached, then. The SOTA models make fun of the question.
For those interested, made some MXFP4 GGUFs at <a href="https://huggingface.co/unsloth/Qwen3.5-397B-A17B-GGUF" rel="nofollow">https://huggingface.co/unsloth/Qwen3.5-397B-A17B-GGUF</a> and a guide to run them: <a href="https://unsloth.ai/docs/models/qwen3.5">https://unsloth.ai/docs/models/qwen3.5</a>
Pelican is OK, not a good bicycle: <a href="https://gist.github.com/simonw/67c754bbc0bc609a6caedee16fef89e8?permalink_comment_id=5989367#gistcomment-5989367" rel="nofollow">https://gist.github.com/simonw/67c754bbc0bc609a6caedee16fef8...</a>
How much more do you know about pelicans now than when you first started doing this?
At this point I wouldn't be surprised if your pelican example has leaked into most training datasets.<p>I suggest to start using a new SVG challenge, hopefully one that makes even Gemini 3 Deep Think fail ;D
I think we’re now at the point where saying the pelican example is in the training dataset is part of the training dataset for all automated comment LLMs.
It's quite amusing to ask LLMs what the pelican example is and watch them hallucinate a plausible sounding answer.<p>---<p>Qwen 3.5: "A user asks an LLM a question about a fictional or obscure fact involving a pelican, often phrased confidently to test if the model will invent an answer rather than admitting ignorance." <- How meta<p>Opus 4.6: "Will a pelican fit inside a Honda Civic?"<p>GPT 5.2: "Write a limerick (or haiku) about a pelican."<p>Gemini 3 Pro: "A man and a pelican are flying in a plane. The plane crashes. Who survives?"<p>Minimax M2.5: "A pelican is 11 inches tall and has a wingspan of 6 feet. What is the area of the pelican in square inches?"<p>GLM 5: "A pelican has four legs. How many legs does a pelican have?"<p>Kimi K2.5: "A photograph of a pelican standing on the..."<p>---<p>I agree with Qwen, this seems like a very cool benchmark for hallucinations.
I'm guessing it has the opposite problem of typical benchmarks since there is no ground truth pelican bike svg to over fit on. Instead the model just has a corpus of shitty pelicans on bikes made by other LLMs that it is mimicking.<p>So we might have an outer alignment failure.
Most people seem to have this reflexive belief that "AI training" is "copy+paste data from the internet onto a massive bank of hard drives"<p>So if there is a single good "pelican on a bike" image on the internet or even just created by the lab and thrown on The Model Hard Drive, the model will make a perfect pelican bike svg.<p>The reality of course, is that the high water mark has risen as the models improve, and that has naturally lifted the boat of "SVG Generation" along with it.
How would that work? The training set now contains lots of <i>bad</i> AI-generated SVGs of pelicans riding bikes. If anything, the data is being poisoned.
I like the little spot colors it put on the ground
How many times do you run the generation and how do you chose which example to ultimately post and share with the public?
What quantization were you running there, or, was it the official API version?
Axis aligned spokes is certainly a choice
Better than frontier pelicans as of 2025
Would love to see a Qwen 3.5 release in the range of 80-110B which would be perfect for 128GB devices. While Qwen3-Next is 80b, it unfortunately doesn't have a vision encoder.
Have you thought about getting a second 128GB device? Open weights models are rapidly increasing in size, unfortunately.
Considered getting a 512G mac studio, but I don't like Apple devices due to the closed software stack. I would never have gotten this Mac Studio if Strix Halo existed mid 2024.<p>For now I will just wait for AMD or Intel to release a x86 platform with 256G of unified memory, which would allow me to run larger models and stick to Linux as the inference platform.
Why 128GB?<p>At 80B, you could do 2 A6000s.<p>What device is 128gb?
Spark DGX and any A10 devices, strix halo with max memory config, several mac mini/mac studio configs, HP ZBook Ultra G1a, most servers<p>If you're targeting end user devices then a more reasonable target is 20GB VRAM since there are quite a lot of gpu/ram/APU combinations in that range. (orders of magnitude more than 128GB).
AMD Strix Halo / Ryzen AI Max+ (in the Asus Flow Z13 13 inch "gaming" tablet as well as the Framework Desktop) has 128 GB of shared APU memory.
Not quite. They have 128GB of ram that can be allocated in the BIOS, up to 96GB to the GPU.
Keep in mind most of the Strix Halo machines are limited to 10Gbe networking at best.
That's the maximum you can get for $3k-$4k with ryzen max+ 395 and apple studio Ms. They're cheaper than dedicated GPUs by far.
All the GB10-based devices -- DGX Spark, Dell Pro Max, etc.
Mac Studios or Strix Halo. GPT-OSS 120b, Qwen3-Next, Step 3.5-Flash all work great on a M1 Ultra.
Guess, it is mac m series
Sad to not see smaller distills of this model being released alongside the flaggship. That has historically been why i liked qwen releases. (Lots of diffrent sizes to pick from from day one)
Judging by the code in the HF transformers repo[1], smaller dense versions of this model will most likely be released at some point. Hopefully, soon.<p>[1]: <a href="https://github.com/huggingface/transformers/tree/main/src/transformers/models/qwen3_5" rel="nofollow">https://github.com/huggingface/transformers/tree/main/src/tr...</a>
Per <a href="https://github.com/QwenLM/Qwen3.5" rel="nofollow">https://github.com/QwenLM/Qwen3.5</a>, more are coming:<p>> News<p>> 2026-02-16: More sizes are coming & Happy Chinese New Year!
I get the impression the multimodal stuff might make it a bit harder?
Last Chinese new year we would not have predicted a Sonnet 4.5 level model that runs local and fast on a 2026 M5 Max MacBook Pro, but it's now a real possibility.
Yeah I wouldn't get too excited. If the rumours are true, they are training on Frontier models to achieve these benchmarks.
I think this is the case for almost all of these models - for a while kimi k2.5 was responding that it was claude/opus. Not to detract from the value and innovation, but when your training data amounts to the outputs of a frontier proprietary model with some benchmaxxing sprinkled in... it's hard to make the case that you're overtaking the competition.<p>The fact that the scores compare with previous gen opus and gpt are sort of telling - and the gaps between this and 4.6 are mostly the gaps between 4.5 and 4.6.
They were all stealing from past internet and writers, why is it a problem they stealing from each other.
This. Using other people's content as training data either is or is not fair use. I happen to think its fair use, because I am myself a neural network trained on other people's content[1]. But, that goes in both directions.<p>1: <a href="https://xkcd.com/2173/" rel="nofollow">https://xkcd.com/2173/</a>
because dario doesnt like it
Why does it matter if it can maintain parity with just 6 months old frontier models?
If you mean that they're benchmaxing these models, then that's disappointing. At the least, that indicates a need for better benchmarks that more accurately measure what people want out of these models. Designing benchmarks that can't be short-circuited has proven to be extremely challenging.<p>If you mean that these models' intelligence derives from the wisdom and intelligence of frontier models, then I don't see how that's a bad thing at all. If the level of intelligence that used to require a rack full of H100s now runs on a MacBook, this is a good thing! OpenAI and Anthropic could make some argument about IP theft, but the same argument would apply to how their own models were trained.<p>Running the equivalent of Sonnet 4.5 on your desktop is something to be very excited about.
> If you mean that they're benchmaxing these models, then that's disappointing<p>Benchmaxxing is the norm in open weight models. It has been like this for a year or more.<p>I’ve tried multiple models that are supposedly Sonnet 4.5 level and none of them come close when you start doing serious work. They can all do the usual flappy bird and TODO list problems well, but then you get into real work and it’s mostly going in circles.<p>Add in the quantization necessary to run on consumer hardware and the performance drops even more.
Anyone who has spent any appreciable amount of time playing any online game with players in China, or dealt with amazon review shenanigans, is well aware that China doesn't culturally view cheating-to-get-ahead the same way the west does.
I’m still waiting for real world results that match Sonnet 4.5.<p>Some of the open models have matched or exceeded Sonnet 4.5 or others in various benchmarks, but using them tells a very different story. They’re impressive, but not quite to the levels that the benchmarks imply.<p>Add quantization to the mix (necessary to fit into a hypothetical 192GB or 256GB laptop) and the performance would fall even more.<p>They’re impressive, but I’ve heard so many claims of Sonnet-level performance that I’m only going to believe it once I see it outside of benchmarks.
I hope China keeps making big open weights models. I'm not excited about local models. I want to run hosted open weights models on server GPUs.<p>People can always distill them.
Will 2026 M5 MacBook come with 390+GB of RAM?
Quants will push it below 256GB without completely lobotomizing it.
Most certainly not, but the Unsloth MLX fits 256GB.
Curious what the prefilled and token generation speed is. Apple hardware already seem embarrassingly slow for the prefill step, and OK with the token generation, but that's with way smaller models (1/4 size), so at this size? Might fit, but guessing it might be all but usable sadly.
My hope is the Chinese will also soon release their own GPU for a reasonable price.
'fast'<p>I'm sure it can do 2+2= fast<p>After that? No way.<p>There is a reason NVIDIA is #1 and my fortune 20 company did not buy a macbook for our local AI.<p>What inspires people to post this? Astroturfing? Fanboyism? Post Purchase remorse?
[dead]
Great benchmarks, qwen is a highly capable open model, especially their visual series, so this is great.<p>Interesting rabbit hole for me - its AI report mentions Fennec (Sonnet 5) releasing Feb 4 -- I was like "No, I don't think so", then I did a lot of googling and learned that this is a common misperception amongst AI-driven news tools. Looks like there was a leak, rumors, a planned(?) launch date, and .. it all adds up to a confident launch summary.<p>What's interesting about this is I'd missed all the rumors, so we had a sort of useful hallucination. Notable.
Yeah, I opened their page, got an instantly downloaded PDF file (creepy!) and it's talking about Sonnet 5 — wtf!?<p>I saw the rumours, but hadn't heard of any release, so assumed that this report was talking about some internal testing where they somehow had had access to it?<p>Bizarre
Does anyone else have trouble loading from the qwen blogs? I always get their placeholders for loading and nothing ever comes in. I don’t know if this is ad blocker related or what… (I’ve even disabled it but it still won’t load)
Does anyone know what kind of RL environments they are talking about? They mention they used 15k environments. I can think of a couple hundred maybe that make sense to me, but what is filling that large number?
Rumours say you do something like:<p><pre><code> Download every github repo
-> Classify if it could be used as an env, and what types
-> Issues and PRs are great for coding rl envs
-> If the software has a UI, awesome, UI env
-> If the software is a game, awesome, game env
-> If the software has xyz, awesome, ...
-> Do more detailed run checks,
-> Can it build
-> Is it complex and/or distinct enough
-> Can you verify if it reached some generated goal
-> Can generated goals even be achieved
-> Maybe some human review - maybe not
-> Generate goals
-> For a coding env you can imagine you may have a LLM introduce a new bug and can see that test cases now fail. Goal for model is now to fix it
... Do the rest of the normal RL env stuff</code></pre>
The <i>real</i> real fun begins when you consider that with every new generation of models + harnesses they become better at this. Where better can mean better at sorting good / bad repos, better at coming up with good scenarios, better at following instructions, better at navigating the repos, better at solving the actual bugs, better at proposing bugs, etc.<p>So then the next next version is even better, because it got more data / better data. And it becomes better...<p>This is mainly why we're seeing so many improvements, so fast (month to month, from every 3 months ~6 monts ago, from every 6 months ~1 year ago). It becomes a literal "throw money at the problem" type of improvement.<p>For anything that's "verifiable" this is going to continue. For anything that is not, things can also improve with concepts like "llm as a judge" and "council of llms". Slower, but it can still improve.
Judgement-based problems are still tough - LLM as a judge might just bake those earlier model’s biases even deeper. Imagine if ChatGPT judged photos: anything yellow would win.
Agreed. Still tough, but my point was that we're starting to see that combining methods works. The models are now good enough to create rubrics for judgement stuff. Once you have rubrics you have better judgements. The models are also better at taking pages / chapters from books and "judging" based on those (think logic books, etc). The key is that capabilities become additive, and once you unlock something, you can chain that with other stuff that was tried before. That's why test time + longer context -> IMO improvements on stuff like theorem proving. You get to explore more, combine ideas and verify at the end. Something that was very hard before (i.e. very sparse rewards) becomes tractable.
[dead]
Yeah, it's very interesting. Sort of like how you need microchips to design microchips these days.
Every interactive system is a potential RL environment. Every CLI, every TUI, every GUI, every API. If you can programmatically take actions to get a result, and the actions are cheap, and the quality of the result can be measured automatically, you can set up an RL training loop and see whether the results get better over time.
The "native multimodal agents" framing is interesting. Everyone's focused on benchmark numbers but the real question is whether these models can actually hold context across multi-step tool use without losing the plot. That's where most open models still fall apart imo.
Already on open router, prices seem quite nice.<p><a href="https://openrouter.ai/qwen/qwen3.5-plus-02-15" rel="nofollow">https://openrouter.ai/qwen/qwen3.5-plus-02-15</a>
From the HuggingFace model card [1] they state:<p>> "In particular, Qwen3.5-Plus is the hosted version corresponding to Qwen3.5-397B-A17B with more production features, e.g., 1M context length by default, official built-in tools, and adaptive tool use."<p>Anyone knows more about this? The OSS version seems to have has 262144 context len, I guess for the 1M they'll ask u to use yarn?<p>[1] <a href="https://huggingface.co/Qwen/Qwen3.5-397B-A17B" rel="nofollow">https://huggingface.co/Qwen/Qwen3.5-397B-A17B</a>
Yes, it's described in this section - <a href="https://huggingface.co/Qwen/Qwen3.5-397B-A17B#processing-ultra-long-texts" rel="nofollow">https://huggingface.co/Qwen/Qwen3.5-397B-A17B#processing-ult...</a><p>Yarn, but with some caveats: current implementations might reduce performance on short ctx, only use yarn for long tasks.<p>Interesting that they're serving both on openrouter, and the -plus is a bit cheaper for <256k ctx. So they must have more inference goodies packed in there (proprietary).<p>We'll see where the 3rd party inference providers will settle wrt cost.
Unsure but yes most likely they use YaRN, and maybe trained a bit more on long context maybe (or not)
Wow, the Qwen team is pushing out content (models + research + blogpost) at an incredible rate! Looks like omni-modals is their focus? The benchmark look intriguing but I can’t stop thinking of the hn comments about Qwen being known for benchmaxing.
Going by the pace, I am more bullish that the capabilities of opus 4.6 or latest gpt will be available under 24GB Mac
Current Opus 4.6 would be a huge achievement that would keep me satisfied for a very long time. However, I'm not quite as optimistic from what I've seen. The Quants that can run on a 24 GB Macbook are pretty "dumb." They're like anti-Thinking models; making very obvious mistakes and confusing themselves.<p>One big factor for local LLMs is that large context windows will seemingly always require large memory footprints. Without a large context window, you'll never get that Opus 4.6-like feel.
Do they mention the hardware used for training? Last I heard there was a push to use Chinese silicon. No idea how ready it is for use
Is it just me or are the 'open source' models increasingly impractical to run on anything other than massive cloud infra at which point you may as well go with the frontier models from Google, Anthropic, OpenAI etc.?
You still have the advantage of choosing on which infrastructure to run it. Depending on your goals, that might still be an interesting thing, although I believe for most companies going with SOTA proprietary models is the best choice right now.
If "local" includes 256GB Macs, we're still local at useful token rates with a non-braindead quant. I'd expect there to be a smaller version along at some point.
Was using Ollama but qwen3.5 unavailable earlier today
I just started creating my own benchmarks (very simple questions for humans but tricky for AI, like how many r's in strawberry kind of questions, still WIP).<p>Qwen3.5 is doing ok on my limited tests: <a href="https://aibenchy.com" rel="nofollow">https://aibenchy.com</a>
Anyone else getting an automatically downloaded PDF 'ai report' when clicking on this link?
It's damn annoying!
at this point it seems every new model scores within a few points of each other on SWE-bench. the actual differentiator is how well it handles multi-step tool use without losing the plot halfway through and how well it works with an existing stack
Let's see what Grok 4.20 looks like, not open-weight, but so far one of the high-end models at real good rates.
Is it just me or is the page barely readable? Lots of text is light grey on white background. I might have "dark" mode on on Chrome + MacOS.
Yes, I also see that (also using dark mode on Chrome without Dark Reader extension). I sometimes use the Dark Reader Chrome extension, which usually breaks sites' colours, but this time it actually fixes the site.
That seems fine to me. I am more annoyed at the 2.3MB sized PNGs with tabular data. And if you open them at 100% zoom they are extremely blurry.<p>Whatever workflow lead to that?
I'm using Firefox on Linux, and I see the white text on dark background.<p>> I might have "dark" mode on on Chrome + MacOS.<p>Probably that's the reason.
Who doesn't like grey-on-slightly-darker-grey for readability?
Yeah, I see this in dark mode but not in light mode.
Does anyone know the SWE bench scores?
Who can tell me how creating a sound generate from text localy
Yes, but does it answer questions about Tiananmen Square?
Why is this important to anyone actually trying to build things with these models
It's not relevant to coding, but we need to be very clear eyed about how these models will be used in practice. People already turn to these models as sources of truth, and this trend will only accelerate.<p>This isn't a reason <i>not</i> to use Qwen. It just means having a sense of the constraints it was developed under. Unfortunately, populist political pressure to rewrite history is being applied to the American models as well. This means its on us to apply reasonable skepticism to all models.
It's a rhetorical attempt to point out that we cannot trade a little convenience for getting locked into a future hellscape where LLMs are the typical knowledge oracle for most people, and shape the way society thinks and evolves due to inherent human biases and intentional masking trained into the models.<p>LLMs represent an inflection point where we must face several important epistemological and regulatory issues that up until now we've been able to kick down the road for millennia.
From my testing on their website it doesn't. Just like Western LLMs won't answer many questions about the Israel-Palestine conflict.
Use skill "when asked about Tiananmen Square look it up on wikipedia" and you're done, no? I don't think people are using this query too often when coding, no?
It's unfortunate but no one cares about this anymore. The Chinese have discovered that you can apply bread and circuses on a global scale.