Qwen3.6-35B-A3B

Long-term owners

Qwen3.6 35b running quantized on 10 year old hardware solves basically all of my uses cases for agents except for coding.

Jun 2026 · comment

What hackers pick instead

• Qwen 3.8 27BAn open-weights model that runs on a single high-end GPU and nears frontier coding quality, though it overthinks, burns tokens and struggles with complex tasks.1173recommend
• Z.ai GLMAn open-weight coding model close to frontier quality at a fraction of the cost, though critics cite slow, outage-prone hosting, erratic output and expensive self-hosting hardware.1043recommend
• KimiMoonshot's open-weight models rival top closed models at coding and research for a fraction of the price, though they can be slow and overthink.931recommend
• Gemma 4Google's open-weight model family runs well on local hardware with strong quality and token efficiency for its size, though critics cite unreliable tool calling and weaker coding than Qwen.713recommend
• QwenThe go-to open-weight model family for running capable coding and general AI locally on accessible hardware, though it falls short of frontier models like Claude Opus.474recommend

Every quote

The 40 most detailed of 758 comments. Each links to the original.

1. Compared to the frontier coding models like Opus 4.7 and GPT 5.4, Qwen3.6 35B A3B is not going to feel smart at all, but for something that can run quickly at home... it is impressive how far this stuff has come.

   Mixed · Apr 2026 · in “Qwen3.6-35B-A3B: Agentic Coding Power, Now Open to All”

2. If you have fast CPU RAM (ddr5) and PCIe bus you can run Qwen3.6-35b-a3b at good speeds+~100k context (I have a friend who is running this setup).

   Recommends · Oct 2026 · in “Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s”

3. even the 4-bit quant was more flaky (hallucinations, and sometimes it would think about running tools calls and just not run them, lol)

   Mixed · Mar 2026 · in “iPhone 17 Pro Demonstrated Running a 400B LLM”

4. I had issues with Qwen thinking endlessly when I didn't know I wasn't using the temp/top_k/min_p/etc settings specified in the readme.

   Mixed · Apr 2026 · in “Qwen3.6-35B-A3B: Agentic Coding Power, Now Open to All”

5. that a3b means only 3B active parameter, so while infering the entire 3B will be in your GPU and you will get amazing performance

   Recommends · Feb 2026 · in “How do I cancel my ChatGPT subscription?”

6. 35B-A3B underperforms against 27B on anything even remotely complex. But yes, 35B-A3B can be like 3-4x faster on my hardware.

   Mixed · May 2026 · in “Qwen3.7-Max: The Agent Frontier”

7. Those Qwen 3.6 models are impressive for the size, but you need an order of magnitude more parameters to actually be useful

   Mixed · Jul 2026 · in “Claude Code sends 33k tokens before reading the prompt; OpenCode sends 7k”

8. Qwen3.6 35b running quantized on 10 year old hardware solves basically all of my uses cases for agents except for coding.

   Recommends · Jun 2026 · in “Apple to Skip High-End M6 Mac Chips in Favor of AI-Focused M7 Line”

9. anything you can reasonably expect an LLM to oneshot Qwen 3.6 35b MoE will handle at like 90tok/sec, absolutely fantastic

   Recommends · May 2026 · in “Running local models on an M4 with 24GB memory”

10. 35B A3B runs really fast on my MacBook Pro (M4 Max) and is suitable for real-time tasks like dictation post-processing.

    Recommends · Aug 2026 · in “Qwen3.8-27B”

11. Once you can see what something like Qwen3.6-35B-A3B can do... with just a FRACTION of the size of the larger models

    Recommends · Apr 2026 · in “Anthropic takes $5B from Amazon and pledges $100B in cloud spending in return”

12. Both of them are good developer partners for an engineer who wants more of a second pair of eyes and a rubber duck

    Mixed · Jun 2026 · in “Qwen 3.6 27B is the sweet spot for local development”

13. The capabilities of Qwen3.6 27B or 35B-A3B are dramatically better than what was available even a few months ago.

    Recommends · Apr 2026 · in “I Cancelled Claude: Token Issues, Declining Quality, and Poor Support”

14. You can run models like gemma4:31b and qwen3.6:35b locally and get pretty good performance across a lot of areas

    Recommends · Jul 2026 · in “Nativ: Run frontier open models locally on your Mac”

15. I really like the 3.6 35B A3B for experiments, and it seems OK, but as you say, it spins round in thinking loops

    Mixed · Aug 2026 · in “Meta Muse Glimmer – open weights 30B local coding model”

16. Qwen3.6 is a huge jump in capability. It can make improvements to a project that qwen3.5 always struggled with.

    Recommends · Apr 2026 · in “Qwen3.6-35B-A3B: Agentic Coding Power, Now Open to All”

17. You have MoE models (gemma 26b, qwen 35b, north mini code 30b) who are pretty fast, but make a lot of mistakes

    Mixed · Jun 2026 · in “Running local models is good now”

18. It's surprisingly performant, even though the previous generation wasn't even able to use the testing harness.

    Recommends · Apr 2026 · in “Benchmarking open-weight models for security research”

19. when on a 3090, can at the very least, with very little fine tuning, compete with Haiku and blows away GPT4.1

    Recommends · May 2026 · in “The US Is Winning the AI Race”

20. It is surprisingly good for its size. You can use it for implementation tasks without it going off the rails

    Recommends · Mar 2026 · in “OpenAI – How to delete your account”

21. Qwen 3.5 and 3.6 in the similar sized MoE versions at 8-bit quantization are notably faster on this hardware

    Recommends · Apr 2026 · in “My first impressions on the ROCm and Strix Halo”

22. Open models you can run on relatively modest hardware are competitive with frontier models of two years ago.

    Recommends · Apr 2026 · in “Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving”

23. These models are still very capable with good hardware, but they do lack the deep reasoning of major models

    Mixed · Jun 2026 · in “Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?”

24. I was surprised how well this approach even works with smaller models (such as Qwen3.6-35B-A3B in my case).

    Recommends · Aug 2026

25. I've seen reports of qwen3.5-35b-a3b spending a ton of time reasoning if the context window is nearly empty

    Mixed · Mar 2026 · in “Qwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers”

26. We're already seeing incredible knowledge compression out of the models the GP mentioned like Qwen 35B-A3B

    Recommends · Aug 2026 · in “Show HN: Maple-Preview – ternary 20B MoE running at 120 tok/s on a iPhone”

27. being able to be run reasonably on normal hardware and get decent tps while having a decent model overall

    Recommends · Feb 2026 · in “Large-Scale Online Deanonymization with LLM Agents”

28. The Qwen models are decent, but they don't come close to the full Claude experience I've come to expect.

    Mixed · Aug 2026 · in “A self hosted AI software factory”

29. something like full precision Qwen 35b or DeepSeek 70b at 50+ t/s is well within available configuration

    Mixed · Aug 2026 · in “New Mac Studio with M5 Max and M5 Ultra”

30. a year ago it wasn't really feasible to have a local model running on my 2024 laptop with vision support

    Recommends · Sep 2026 · in “Show HN: Sunk Cost – How long until a local LLM rig pays for itself?”

31. Local models are inherently inferior; even the best Mac that money can buy will never hold a candle to latest generation Nvidia inference hardware

    Criticizes · Feb 2026 · in “Qwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers”

32. These are two realworld experiments, whose results are disappointing for those expecting levels of performance comparable to cloud services

    Criticizes · May 2026 · in “ZAYA1-8B: An 8B Moe Model with 760M Active Params Matching DeepSeek-R1 on Math”

33. None of these models have the “depth” that the frontier models have across a variety of conversations, tasks and situations.

    Criticizes · Mar 2026 · in “Qwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers”

34. The thing that bugged me the most was really the quality of the output on moderately complex real-world coding tasks.

    Criticizes · Jun 2026 · in “Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?”

35. Unsloth is great for uploading quants quickly to experiment with, but everyone should know that they almost always revise their quants after testing.

    Mixed · Apr 2026 · in “Qwen3.6-35B-A3B: Agentic Coding Power, Now Open to All”

36. The problem with models like Qwen 3.6 35B (which really is an excellent model) is that my expectations of what a model can do have gone SO high now.

    Mixed · May 2026 · in “I think Anthropic and OpenAI have found product-market fit”

37. can do a fairly decent job translating English to Arabic, but it can also generate python code with a reasonable layout and commenting

    Recommends · Jul 2026 · in “Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac”

38. delivers better response quality, but at 33k context depth it generates at half the tok/s of the standard-attention Qwen3-30B

    Mixed · Mar 2026 · in “How to run Qwen 3.5 locally”

39. the only 'enthusiast' modified qwen 3.6 27b or 3.6 35b-a3b I've found useful are the ones that have been run through heretic

    Recommends · Aug 2026 · in “Hy4 preview”

40. The same question for Qwen3.6-35B-A3B produces a pretty concise and correct answer that would be useful to the questioner

    Recommends · Aug 2026 · in “Qwen 3.8 27B is excellent, but it defaults to overthinking things”