I cant wait to see some smaller sizes. I would love to run some sort of coding centric agent on a local TPU or GPU instead of having to pay, even if it's slower.
Qwen3.6-35B-A3B: Agentic coding power, now open to all
311–320 of 563 posts
Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all
#312I've been running this on my laptop with the Unsloth 20.9GB GGUF in LM Studio: https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF/blob/mai... It drew a better pelican riding a bicycle than Opus 4.7 did! https://simonwillison.net/2026/Apr/16/qwen-beats-opus/
I wonder when pelican riding a bicycle will be useless as an evaluation task. The point was that it was something weird nobody had ever really thought about before, not in the benchmarks or even something a team would run internally. But now I'd bet internally this is one of the new Shirley Cards.
Simon has an article on this
https://simonwillison.net/2025/Nov/13/training-for-pelicans-...
Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all
#313I've been running this on my laptop with the Unsloth 20.9GB GGUF in LM Studio: https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF/blob/mai... It drew a better pelican riding a bicycle than Opus 4.7 did! https://simonwillison.net/2026/Apr/16/qwen-beats-opus/
The flamingo on Qwen's unicycle is sitting on the tire, not the seat. That wins because of sunglasses?
Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all
#314I've been running this on my laptop with the Unsloth 20.9GB GGUF in LM Studio: https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF/blob/mai... It drew a better pelican riding a bicycle than Opus 4.7 did! https://simonwillison.net/2026/Apr/16/qwen-beats-opus/
I wonder when pelican riding a bicycle will be useless as an evaluation task. The point was that it was something weird nobody had ever really thought about before, not in the benchmarks or even something a team would run internally. But now I'd bet internally this is one of the new Shirley Cards.
I mean look at the result where he asked about a unicycle - the model couldn't even keep the spokes inside the wheels - would be rudimentary if it "learned" what it means to draw a bicycle wheel and could transfer that to unicycle.
Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all
#315Looks like they compare only to open models, unfortunately. As I am using mostly the non-open models, I have no idea what these numbers mean.
[dead]
Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all
#316Already quantized/converted into a sane format by Unsloth: https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF
Why doesn't Qwen itself release the quantized model? My impression is that quantization is a highly nontrivial process that can degrade the model in non-obvious ways, thus its best handled by people who actually built the model, otherwise the results might be disappointing. Users of the quantized model might be even made to think that the model sucks because the quantized version does.
Quantization is an extraordinarily trivial process. Especially if you're doing it with llama.cpp (which unsloth obviously does).
Qwen did release an fp8 version, which is a quantized version.
Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all
#317Already quantized/converted into a sane format by Unsloth: https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF
There's absolutely nothing wrong it insane with a safetensors file. It might be less convenient than a single file gguf. But that's just laziness not insanity
Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all
#318I recall a Qwen exec posted a public poll on Twitter, asking which model from Qwen3.6 you want to see open-sourced; and the 27b variant was by far the most popular choice. Not sure why they ignored it lol.
The 27B model is dense. Releasing a dense model first would be terrible marketing, whereas 35A3B is a lot smarter and more quick-witted by comparison!
That makes no sense. If you were just going to release the "more hype-able because it's quicker" model then why have a a poll.
Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all
#319Earlier quoted context omitted.
This is why I'd like to see a lot more focus on batched inference with lower-end hardware. If you just do a tiny amount of tok/day and can wait for the answer to be computed overnight or so, you don't really need top-of-the-line hardware even for SOTA results.
> If you just do a tiny amount of tok/day and can wait for the answer to be computed overnight or so But they can't? The usage pattern is the polar opposite. Most people running these models locally just ask a few questions to it throughout the day. They want the answers now, or at least within a minute.
If you want the answer right now, that alone ups your compute needs to the point where you're probably better off just using a free hosted-AI service. Unless the prompt is trivial enough that it can be answered quickly by a tiny local model.
Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all
#320I've been running this on my laptop with the Unsloth 20.9GB GGUF in LM Studio: https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF/blob/mai... It drew a better pelican riding a bicycle than Opus 4.7 did! https://simonwillison.net/2026/Apr/16/qwen-beats-opus/
I've had some really gnarly SVGs from Claude. Here's what I got after many iterations trying to draw a hand: https://imgur.com/a/X4Jqius
Probably because all the training material of humans drawing hands are garbage haha.