Qwen 3.6
Alibaba's open-weight LLM family, including 27B dense and 35B-A3B mixture-of-experts models.
Liked as a capable model that runs locally on consumer GPUs and Macs1234 and is many commenters' daily driver for coding and everyday tasks.56789 The most common caveat is that it falls short of frontier models on coherence and harder coding work1011121314, and some report runaway thinking loops and failed tool calls.15161718
Why people recommend it
Daily driver · 3
Many use it as their main model for coding and general work.
“Qwen3.6-35B is my daily driver for AI, and what convinced me to cancel my Claude subscription back in April.”
“3.6 27B was easily a daily driver with only an occasional need to pop up to larger models for planning.”
Show 1 more
“I have been very blown away by how well it works.”
Consumer hardware · 3
Runs usefully on modest GPUs and Macs, including 8GB VRAM cards and 16GB laptops.
“It's a very capable model, not as smart as a frontier model but it is still very useful.”
“I am running Qwen3.6-35B-A3B on my 16GB M3 and get 7-8 tokens/sec”
Show 1 more
“a single 3090 already provides a great experience at a reasonable quant and context size”
Speed · 3
The 35B-A3B MoE variant in particular reaches high token rates locally.
“The moe works well in Librechat combined with my local search/Web retrieval system.”
“Been using Qwen3.6-35B @ Q_8 as my main driver and it’s been great with 60 TPS for generation.”
Show 1 more
“it's both much faster and much more capable than Haiku”
Replaces paid APIs · 3
Some commenters dropped Claude subscriptions or use it to save cloud tokens.
“Qwen3.6-35B is my daily driver for AI, and what convinced me to cancel my Claude subscription back in April.”
“I use it for personal tasks that are just not worth the claude tokens”
Show 1 more
“been happy throwing Qwen3.6 on my old MBP”
Agents and tools · 3
Works well in agent setups, eagerly using web search and orchestrating other agents.
“That's why I like qwen3.6 27B, it has 0 ego, it knows that it doesn't have complete world knowledge”
“Then I have Qwen3.6-35B-A3B as the "orchestrator" agent (it's just what I use for my daily driver)”
Show 1 more
“It's the first local model that's actually working for me in a coding agent I've tried.”
Caveats and dissent
Below frontier · 3
Commenters find it less coherent and capable than frontier models for serious development.
“the coherence was really not great compared to frontier”
“for software development it is unsurprisingly a dramatic downgrade from the Opus I use as my daily driver.”
Show 1 more
“it can't get past a simple coding test that 5.5 passes with light reasoning”
Thinking loops · 2
It can spin in long thinking loops that some cap with a proxy.
“My final 35B A3B tip -- use an extremely simple harness.”
Tool-call glitches · 3
Some see failed tool calls, mid-thinking cutoffs or parsing problems, partly fixed by improved templates.
“I tried Qwen 3.6 27B, but I can only get around 10 tokens per second, but it's painfully slow, and it often fails during `write_file` tool calls”
“35B is pretty good”
Show 1 more
“I use Qwen 3.6 a fair amount using his template and it fixes some issues I saw with the upstream versions.”
Needs precise specs · 3
It struggles with vague specs and needs detailed upfront guidance or split sessions.
“This is a fun little model that's smart enough to do debugging, refactoring, and implementing "clean" specs”
“I also "evolved" into 27b (q8 unsloth)”
Show 1 more
“I run Qwen3.6 on RTX4090, and it does amazing job for the most parts.”
Laptop heat · 1
Running it on a laptop during heavy work causes heat and fan noise.
“I love my MacBook Pro M5 128GB RAM and I love qwen3.6.”
Often compared with
2 moved from Qwen 3.5 to Qwen 3.6
1 moved from Claude to Qwen 3.6
Who recommends it in their own words
Threads where it came up
- Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding? 26 mentions
- Running local models is good now 13 mentions
- Qwen 3.6 27B is the sweet spot for local development 12 mentions
- Qwen3.8-Max: A New Bar for Coding and Cowork 6 mentions
- Qwen 3.8 27B is excellent, but it defaults to overthinking things 3 mentions
- Jamesob's guide to running SOTA LLMs locally 3 mentions
Also written as Qwen 3.6 27B, Qwen 3.6, Qwen3.6-27B, Qwen3.6 27B. Data: qwen-3-6.json. Quotes are verbatim; the verdict and theme labels are written by a model from these comments, and every clause links to its sources.
All comments 49
“I have repurposed my 6 year old 32gb M1 MacBook Pro to run Qwen 3.6 35b a3b”
“the coherence was really not great compared to frontier”
“trying to run models like Qwen3.6 35B A3B”
“Local model is Qwen3.6 35B-A3B and Qwen3.8 27B UD.”
“Qwen 3.6 a3b makes more mistakes than DeepSeek but it's free.”
“it's both much faster and much more capable than Haiku”
“I find them quite useful!”
“Qwen3.6-35B-A3B (speed) and Qwen3.8-27B (intelligence, but slow)”
“running Qwen 3.6 35b a3b sipping 10w when generating tokens is good enough for a lot of my guide-coding work”
“Qwen3.6-35B-A3B (speed)”
“I also use DeepSeek V4 flash and a Qwen 3.6 MoE for local stuff.”
“honestly don’t even bother using qwen3.6-35b-a3b now unless i need uncensored tasks finished, mostly reverse engineering”
“the smaller Gemma 4 models or qwen3.6:9b would probably run OK on that”
“I run Qwen3.6-35B-A3B-MTP (fast!)”
“I am running Qwen3.6-35B-A3B on my 16GB M3 and get 7-8 tokens/sec”
“I've used qwen3.6 and it's not as straightforward as”
“definitely better than Qwen3.6 35b-a3b which was my go-to before”
“It's a very capable model, not as smart as a frontier model but it is still very useful.”
“but those do exit eventually, unlike with Qwen 3.6.”
“I've only benchmarked 27B across Q3-6 quants using lm-eval.”
“I used Qwen 3.6 27B extensively (>1B tokens)”
“3.6 27B was easily a daily driver with only an occasional need to pop up to larger models for planning.”
“This was mainly put together to rein in Qwen3.6-27B”
“I use Qwen 3.6 a fair amount using his template and it fixes some issues I saw with the upstream versions.”
“Use 3.6 27b as a daily driver for months with charmbracelet crush.”
“the best models you can run at home are probably Qwen 3.6 27B and Gemma 4 31B”
“I use qwen3.6 27b for personal projects and I think it works pretty great.”
“My final 35B A3B tip -- use an extremely simple harness.”
“I love the 3.6 version”
“The moe works well in Librechat combined with my local search/Web retrieval system.”
“i used to play around with qwen 3.6 35b a3b with 17/tps”
“Then I have Qwen3.6-35B-A3B as the "orchestrator" agent (it's just what I use for my daily driver)”
“start with Qwen3.6, Google's Gemma4, OpenAI's gpt-oss-20b or Nvidia's Nemotron”
“local Qwen3.6-35B A3B”
“I use it for personal tasks that are just not worth the claude tokens”
“Sure a lot of this could be done without AI, but it's certainly quicker and easier”
“I used the qwen 3.6 models and convinced myself to go ahead and get a gpu.”
“I also "evolved" into 27b (q8 unsloth)”
“Qwen3.6-35B is my daily driver for AI, and what convinced me to cancel my Claude subscription back in April.”
“I have qwen 3.6 27b in an rtx 6000 pro as well so I use that as a workhorse in pi”
“"I love hallucinated garbage": Sonnet, Qwen 3.6, GPT 5.4 mini, GPT 5.3 Codex”
“a single 3090 already provides a great experience at a reasonable quant and context size”
“I tried Qwen 3.6 27B, but I can only get around 10 tokens per second, but it's painfully slow, and it often fails during `write_file` tool calls”
“35B is pretty good”
“for software development it is unsurprisingly a dramatic downgrade from the Opus I use as my daily driver.”
“it can't get past a simple coding test that 5.5 passes with light reasoning”
“A 32gb card does run it nicely.”
“I have been very blown away by how well it works.”
“I use qwen3.6:27b as a daily driver and I am heavily impressed with it.”