Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Qwen3.6-35B-A3B: Agentic coding power, now open to all

qwen.ai

Post reply on HN
Print view View original

301–310 of 563 posts

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#301

Post by simonw »

cyclopeanutopia wrote: post #290
simonw wrote: post #270

I'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/

But that you also gave a win to Qwen on flamingo is pretty outrageous! :) Tthe right one looks much better, plus adding sunglasses without prompting is not that great. Hopefully it won't add some backdoor to the generated code without asking. ;)

I love how the Chinese models often have an unprompted predilection to add flair.

GLM-5.1 added a sparkling earring to a north Virginia opossum the other day and I was delighted: https://simonwillison.net/2026/Apr/7/glm-51/

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#302

Post by jubilanti »

simonw wrote: post #270

I'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.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#303

Post by deaux »

lkjdsklf wrote: post #135
aliljet wrote: post #73

I'm broadly curious how people are using these local models. Literally, how are they attaching harnesses to this and finding more value than just renting tokens from Anthropic of OpenAI?

The people i know that use local models just end up with both. The local models don’t really compete with the flagship labs for most tasks But there are things you may not want to send to them for privacy reasons or tasks where you don’t want to use tokens from your plan with whichever lab. Things like openclaw use a ton of tokens and most of the time the local models are totally fine for it (assuming you find it use…

The open weights models absolutely compete with flagship labs for most tasks. OpenAI and Anthropic's "cheap tier" models are completely uncompetitive with them for "quality / $" and it's not close. Google is the only one who has remained competitive in the Unless you have a corporate lock-in/compliance need, there has been no reason to use Haiku or GPT mini/nano/etc over open weights models for a long time now.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#304

Post by bityard »

torginus wrote: post #279
homebrewer wrote: post #18

Already 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.

Model developers release open-weight models for all sorts of reasons, but the most common reason is to share their work with the greater AI research community. Sure, they might allow or even encourage personal and commercial use of the model, but they don't necessarily want to be responsible for end-user support.

An imperfect analogy might be the Linux kernel. Linus publishes official releases as a tagged source tree but most people who use Linux run a kernel that has been tweaked, built, and packaged by someone else.

That said, models often DO come from the factory in multiple quants. Here's the FP8 quant for Qwen3.6 for example: https://huggingface.co/Qwen/Qwen3.6-35B-A3B-FP8

Unsloth and other organizations produce a wider variety of quants than upstream to fit a wider variety of hardware, and so end users can make their own size/quality trade-offs as needed.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#305

Post by gunalx »

tommy_axle wrote: post #184
palmotea wrote: post #67

Earlier quoted context omitted.

How much VRAM does it need? I haven't run a local model yet, but I did recently pick up a 16GB GPU, before they were discontinued.

Pick a decent quant (4-6KM) then use llama-fit-params and try it yourself to see if it's giving you what you need.

I habe found llama-fit sometimes just selects a way to conservative load with VRAM to spare.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#306

Post by txtsd »

Ladioss wrote: post #175
txtsd wrote: post #54

Earlier quoted context omitted.

So I can use this in claude code with `ollama run claude`?

More like `ollama launch claude --model qwen3.6:latest` Also you need to check your context size, Ollama default to 4K if <24 Gb of VRAM and you need 64K minimum if you want claude to be able to at least lift a finger.

I only have 16GB VRAM, and my system uses ~4GB from that. What are my options? I got this one: `Qwen3.6-35B-A3B-UD-IQ2_XXS.gguf`

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#307

Post by gunalx »

trvz wrote: post #84
palmotea wrote: post #67

Earlier quoted context omitted.

How much VRAM does it need? I haven't run a local model yet, but I did recently pick up a 16GB GPU, before they were discontinued.

If you have to ask then your GPU is too small. With 16 GB you'll be only able to run a very compressed variant with noticable quality loss.

Running q3 xss with full and quantizised context as options on a 16gb gpu and still has pretty decent quality and fitting fine with up to 64k context.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#308

Post by canarias_mate »

Aurornis wrote: post #176
homebrewer wrote: post #18

Already quantized/converted into a sane format by Unsloth: https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF

Unsloth is great for uploading quants quickly to experiment with, but everyone should know that they almost always revise their quants after testing. If you download the release day quants with a tool that doesn’t automatically check HF for new versions you should check back again in a week to look for updated versions. Some times the launch day quantizations have major problems which leads to early adopters dismissi…

[flagged]

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#309

Post by coppsilgold »

ndriscoll wrote: post #99
smrtinsert wrote: post #38

Earlier quoted context omitted.

How true is this? How does a regulated industry confirm the model itself wasn't trained with malicious intent?

Why would it matter if the model is trained with malicious intent? It's a pure function. The harness controls security policies.

Much like a developer can insert a backdoor as a "bug" so can an LLM that was trained to do it.

One way you could probably do it is by identifying a commonly used library that can be misused in a way that would allow some kind of time-of-check to time-of-use (TOCTOU) exploit. Then you train the LLM to use the library incorrectly in this way.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#310

Post by vidarh »

pzo wrote: post #181
vidarh wrote: post #10

Earlier quoted context omitted.

The will not measure up. Notice they're comparing it to Gemma, Google's open weight model, not to Gemini, Sonnet, or GPT. That's fine - this is a tiny model. If you want something closer to the frontier models, Qwen3.6-Plus (not open) is doing quite well[1] (I've not tested it extensively personally): https://qwen.ai/blog?id=qwen3.6

on the bright side also worth to keep in mind those tiny models are better than GPT 4.0, 4.1 GPT4o that we used to enjoy less than 2 years ago [1] [1] https://artificialanalysis.ai/?models=gpt-5-4%2Cgpt-oss-120b...

They're absolutely worth using for the right tasks. It's hard to go back to GPT4 level for everything (for me at least), but there's plenty of stuff they are smart enough for.
Post reply on HN