Qwen 3.5
Alibaba's family of open-weight large language models, from 0.8B up to 397B parameters.
Liked for running capably on consumer hardware1234, with small variants handling narrow tasks well and larger ones doing workable local coding.5678 The most common complaint is looping and overthinking unless sampling settings and quants are right91011, and commenters say it still trails frontier models like Opus.121314
Why people recommend it
Consumer hardware · 3
Commenters report usable speeds on 12–16GB GPUs, Macs, and even single-board computers.
“it is by far the most capable model that could fit on my machine”
“I've cracked the optimal configuration for Qwen3.5-35B-A3B on consumer 16GB GPUs.”
Show 1 more
“I'm running Qwen3.5:0.8b locally on an Orangepi Zero 2w using llama.cpp, runs just fine on cpu only.”
Small-task models · 3
The tiny 0.8B–9B variants are described as surprisingly good at narrow, well-scoped jobs.
“qwen3.5 4b and 9b have both been surprisingly good for small tasks.”
“I have had really good results from Qwen 3.5 1B, 4B, etc.”
Show 1 more
“It works surprisingly well really.”
Local coding · 3
Users run it with coding agents and harnesses and say it writes half-decent code and clears tasks without getting stuck.
“I've been using qwen3.5 (122b) with claude code for months, and it's definitely dumber than sonnet/opus, but it works through things reasonably well”
“Qwen 3.5 9B with Opencode was much faster and actually able to work through a majority of the lint warnings without getting stuck”
Show 1 more
“I used Qwen3.5-35B on Mac M1 Max with oMLX. It wrote 98% of the code”
Near-cloud quality · 3
Some found its output comparable to Claude Sonnet or Opus for specific tasks such as summaries and one-shot scenes.
“The quality of the qwen output was comparable”
“I have settled on the qwen3.5 family after testing most of the local usable models.”
Show 1 more
“Its by far better than Opus.”
Data control · 2
Running it locally or on owned hardware keeps compute and data under the user's control.
“This works for us and will work for years to come.”
“Speed is pretty acceptable for the size: ~5s TTFT and ~100 t/s generation, all running on a Macbook M5.”
OCR and vision · 2
The small models handle degraded text and image input better than classic OCR tools, according to commenters.
“Modern VLMs put classic FineReader to shame for processing low-resolution/degraded/non-standard text.”
“I've cracked the optimal configuration for Qwen3.5-35B-A3B on consumer 16GB GPUs.”
Caveats and dissent
Looping · 3
It can loop or burn many thinking tokens, especially with early quants or wrong sampling parameters.
“Qwen 3.5 Q4 went into some kind of loop where it fixed whatever was broken on the previous iteration only to have it broken some other way.”
“the models are still quite capable when configured that way.”
Show 1 more
“I love how literal Qwen 3.5 was”
Below frontier · 3
Users call it dumber than Sonnet/Opus and say it needs harness tuning to get close.
“I've been using qwen3.5 (122b) with claude code for months, and it's definitely dumber than sonnet/opus, but it works through things reasonably well”
“it's dumb as a stump but still smart enough to get some useful work done”
Show 1 more
“I use mostly Qwen3.5:122b-a10b but its the same family and the harness I've developed for it has made a huge difference”
Coding mistakes · 1
One user switched back to Gemma 4 after Qwen kept making mistakes that needed fixing.
“My orchestration model kept having to send the Qwen model back to fix mistakes that Gemma wouldn't have made.”
Censorship · 1
The regular 9B model sometimes refuses or gives sanitized answers on politically sensitive topics.
“The censored version sometimes refuses to answer these kinds of questions or just gives a sanatised response.”
Tiny-model inconsistency · 1
The 0.8B and 2B models are described as very inconsistent at following instructions.
“Qwen 3.5 0.8B and 2B are smart but very inconsistent for instruction following.”
Speed and context · 3
Some users find local runs slower than cloud models, and the 9B is said to work best with small contexts.
“The quality of the qwen output was comparable”
“I use it with Qwen 3.5 running locally when my daily limits run out”
Show 1 more
“Qwen 3.5:9b is quite good as long as problems/context remains smaller.”
Often compared with
2 moved from Qwen 3.5 to Qwen 3.6
1 moved from GPT-OSS to Qwen 3.5
1 moved from Claude Opus to Qwen 3.5
2 moved from Qwen 3.5 to GemmaPi6 comments mention both
Who recommends it in their own words
Threads where it came up
- Qwen3.6-35B-A3B: Agentic Coding Power, Now Open to All 5 mentions
- Running local models is good now 3 mentions
- VibeThinker: 3B param model that beats Opus 4.5 on reasoning with novel SFT+GRPO 2 mentions
- Local Qwen isn't a worse Opus, it's a different tool 2 mentions
- Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding? 2 mentions
- Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving 2 mentions
Also written as Qwen3.5, Qwen 3.5 9B, Qwen3.5 9B, Qwen3.5-122B-A10B. Data: qwen-3-5.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 44
“made a 500MB model for CPU based off qwen 3.5 0.8B”
“Qwen 3.5 Q4 went into some kind of loop where it fixed whatever was broken on the previous iteration only to have it broken some other way.”
“qwen3.5 4b and 9b have both been surprisingly good for small tasks.”
“using a much weaker model e.g. Qwen 3.5 4B”
“Also serves Qwen 3.5 9B when I use my MacBook Air.”
“qwen 3.5 9b q8_0 does about 45”
“My stack is basically deer-flow with Qwen3.5-122B-A10B”
“On Qwen 3.5-9b I get at most 10tk/s.”
“I love how literal Qwen 3.5 was”
“I use mostly Qwen3.5:122b-a10b but its the same family and the harness I've developed for it has made a huge difference”
“my Uni is self hosting Kimi K2.7 and Qwen 3.5”
“Bonsai vs Qwen (quick) Benchmark”
“Qwen_Qwen3.5-27B-Q4_K_M.gguf”
“I have had really good results from Qwen 3.5 1B, 4B, etc.”
“Qwen 3.5 0.8B and 2B are smart but very inconsistent for instruction following.”
“Modern VLMs put classic FineReader to shame for processing low-resolution/degraded/non-standard text.”
“I recommend giving Qwen 3.5 122B-A10B a shot.”
“I use really small models, like Qwen 3.5 0.8B to 9B - no tool calling, no MCP, no skills, nothing.”
“Even qwen3.5 9B is mostly search-eager (but given the size, it's weaker on reasoning”
“Its by far better than Opus.”
“I've been running qwen3-5-9b-q4-k-m and qwen3-6-27b-q6-k simultaneously on an Intel Arc Pro B70 with a lot of success.”
“I moved from it to Qwen3.6-27B (both Q8), and I prefer 3.6-27B”
“qwen35 27B Q4_K - Medium”
“Speed is pretty acceptable for the size: ~5s TTFT and ~100 t/s generation, all running on a Macbook M5.”
“Qwen3.5 9B for fast inference on simple things”
“Qwen3.5-122B-A10B-FP8, Kimi-K2.6.”
“last I used it was with Qwen 3.5”
“It works surprisingly well really.”
“currently I am trying with Qwen 3.5 0.8B”
“Models like the qwen variants, minimax, k2.5, glm models are available by American providers like AWS at a much cheaper price”
“it's dumb as a stump but still smart enough to get some useful work done”
“I use LM studio and qwen3.5 35B”
“Qwen 3.5:9b is quite good as long as problems/context remains smaller.”
“Qwen3.5-122B-A10B currently in evals for agentic coding.”
“I've started getting better results once I switched from gemma 4 26B to qwen3.5 9B”
“This works for us and will work for years to come.”
“Qwen 3.5, Gemma 4, even gpt-oss-120b are solid at reasonable quants if they have the right data.”
“Qwen 3.5 9B with Opencode was much faster and actually able to work through a majority of the lint warnings without getting stuck”
“I've been using qwen3.5 (122b) with claude code for months, and it's definitely dumber than sonnet/opus, but it works through things reasonably well”
“The quality of the qwen output was comparable”
“it is by far the most capable model that could fit on my machine”
“the models are still quite capable when configured that way.”
“I've cracked the optimal configuration for Qwen3.5-35B-A3B on consumer 16GB GPUs.”
“checkout Qwen3.5 the 27B and 35B variants”