Qwen 3.8 27B

Long-term owners

27B is a cracking, little model. It is the first small model that I, as someone with 30 years of experience, can finally say is good enough

Oct 2026 · comment

What hackers pick instead

• 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
• Qwen3.6-35B-A3BA fast mixture-of-experts open model that runs well on consumer hardware and is surprisingly capable for its size, though it falls well short of frontier models on complex tasks.373recommend

Every quote

The 40 most detailed of 2985 comments. Each links to the original.

1. the smaller parameter models are not really comparable to Opus when it comes to making intelligent decisions about greyer areas of good software architecture

   Mixed · Aug 2026 · in “GLM-5.3-Flash”

2. it FEELS like opus 4.5, i gave it an image and a broad overview of what I wanted it to build, and it built the whole thing from beginning to end.

   Recommends · Aug 2026 · in “Qwen 3.8 27B is out: open weights, best local dense model yet”

3. until I can ask their models what happened during the student protests in 1989 I'm just not able to trust these models with anything of substance

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

4. Many of them (K2.5 the Minimax, the GLM models, the Qwen 3.6 models) are about as capable as frontier US models from about 4 months ago or so.

   Recommends · Apr 2026 · in “Building a SaaS in 2026 Using Only EU Infrastructure”

5. The recent Qwen 3.5 9B model is quite good, though it did trip all over itself to avoid telling me what happened on Tiananmen Square in 1989.

   Mixed · Mar 2026 · in “Ollama is now powered by MLX on Apple Silicon in preview”

6. It's definitely burning through a lot of tokens to ruminate about aspects of tasks that earlier models get done better seemingly on memory.

   Mixed · Aug 2026 · in “Qwen 3.8 27B is out: open weights, best local dense model yet”

7. 27B is a cracking, little model. It is the first small model that I, as someone with 30 years of experience, can finally say is good enough

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

8. If you just need to implement Python code from an existing plan Qwen is your choice but it has problem succeeding with more complex tasks

   Mixed · Jun 2026 · in “Unsloth GLM-5.2 – How to Run Locally”

9. Today, you should be able to run Qwen 3.8 - 27B amazingly well on either which is giving comparable performance to 5.6 Luna on SWE Bench.

   Recommends · Aug 2026 · in “Apple introduces M6 and M5 Ultra”

10. I find the assertion that the model's coding performance surpasses their own flagship 397B model from last generation fairly convincing.

    Recommends · Apr 2026 · in “Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model”

11. it's not great at diagnosing issues in the code, but if I can clearly articulate a test suite or boilerplate refactoring it works fine

    Mixed · Jun 2026

12. I tried again with the 2.4T model and it still ruminates to death, but with thinking turned off, it generated genuinely useful advice.

    Mixed · Aug 2026 · in “I gave Qwen 3.8 27B a reverse-engineering job and it finished in 30 minutes”

13. It feels very snappy and unlike my experience with running 27B models the performance stays pretty flat even as the context increases

    Recommends · Sep 2026 · in “Shapelearn Qwen 3.8 27B (13.1 GB VRAM)”

14. It's a good model, but it hallucinates way more than the frontier models and often diverges even from the very explicit instructions.

    Mixed · Oct 2026 · in “Anthropic conjures threat that doesn't exist yet to stoke fears of AI apocalypse”

15. Very impressive given the cost, especially since running locally you just have one time acquisition of hardware and then electricity.

    Recommends · Oct 2026 · in “Five Qwen3.8-27B models match Fable 5 Using GVS5H Harness”

16. Not very fast at 20 tps (8-bit quant for total memory usage around 30 GB), but enough to feel that I have a true local coding buddy.

    Mixed · Aug 2026 · in “Qwen 3.8 27B is out: open weights, best local dense model yet”

17. The model (Qwen 3.8 27b) that I run on my own hardware on a single, high end gaming GPU, is often compared to the state of the art

    Recommends · Aug 2026 · in “Changes to Sourcehut's terms of service regarding LLMs”

18. tiny, highly cached, fast models like Qwen's 27b/a3b range or DeepSeek flash are highly capable and relatively inexpensive to run

    Recommends · Aug 2026 · in “Coding expertise is going to collapse from AI reliance”

19. the locally hosted Qwen is good enough that looking at a given piece of work output I couldn't tell you which model was behind it

    Recommends · Aug 2026 · in “GLM-5.3-Flash”

20. Qwen 3.6 27b is already a viable default. I'm running it on a single 7900 XTX right now for Go development with pi. It's great.

    Recommends · Aug 2026 · in “Qwen3.8 Max now ranked as the best overall model by agentic index”

21. Qwen 3.8 27B is only the second local model after Gemma 4 that managed to correctly reason through one of my private benchmarks

    Mixed · Aug 2026 · in “Qwen 3.8 27B is out: open weights, best local dense model yet”

22. Qwen3.8 27B does not seem to be good enough to implement the simulator with all its intricacies and tiny subtleties correctly

    Mixed · Sep 2026 · in “Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses”

23. With local inference on pretty decent local models we have nowadays (Qwen-3.5 and better) it's not much of a concern anymore.

    Recommends · Apr 2026 · in “Instant 1.0, a backend for AI-coded apps”

24. It understands code well enough to get 80% of the way to a good design, and it can fully implement a well-specified feature.

    Mixed · May 2026 · in “Microsoft Drops Claude Code After Budget Overrun”

25. These speeds are somewhat faster than what I've seen from commercial SOTA models, they're plenty fast for many applications.

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

26. With thinking on it just ruminates forever and with thinking off it gives obviously bad borderline hallucinating advice.

    Mixed · Aug 2026 · in “I gave Qwen 3.8 27B a reverse-engineering job and it finished in 30 minutes”

27. This is why I have switched nearly all of my personal coding experiments over to Qwen3.6 27B.

    Recommends · May 2026 · in “Microsoft Drops Claude Code After Budget Overrun”

28. The earlier Qwen models freely talk about Tiananmen square when not served from China.

    Recommends · Jan 2026 · in “Qwen3-Max-Thinking”

29. How convenient of them to compare themselves to the last generation Opus and GPT models to make their model look better than it really is.

    Criticizes · Apr 2026 · in “Qwen3.6-Plus: Towards Real World Agents”

30. It doubts itself so much and questions its every decision to the point that it'll literally blow the entire context on thinking alone

    Criticizes · Aug 2026 · in “Fable and the end of the free lunch”

31. Nowhere near the power of ChatGPT 5.4 Pro imho... thought for maybe 15 seconds on a problem that pro would have spend 15 minutes on

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

32. Ridiculous amount of "Wait, actually.." which might help for complex coding tasks but makes it unbearable for general purpose use.

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

33. This looks like it's coming from a separate "safety mechanism". Remains to be seen how much censorship is baked into the weights.

    Criticizes · Jan 2026 · in “Qwen3-Max-Thinking”

34. The training of Qwen's 27B parameter model showed that by structuring model training from fundamentals to more difficult topics they were able to drastically reduce the number of parameters needed.

    Recommends · Aug 2026 · in “GLM-5.3-Flash”

35. I did test it on controversial topics that I already know various sides of the argument and I could see it worked well to give a well-rounded exploration of the issue.

    Recommends · Jan 2026 · in “Qwen3-Max-Thinking”

36. I keep finding more and more usecases for Q3.6 27b (same league) and the best performance is, when answers to my question is already in the context.

    Recommends · Jun 2026 · in “RTX 5080 and RTX 3090 Setup: 80 Tok/s on Qwen 3.6 27B Q8”

37. I'm not seeing that with that exact quantization from Unsloth, so far. it's taking a long time, but I don't think it's doing anything pathological.

    Mixed · Aug 2026 · in “Qwen 3.8 27B is out: open weights, best local dense model yet”

38. Qwen seems better at one-shotting things based on vague prompts to an acceptable degree, but thats literally not what I use these things for!

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

39. Same goes for the full Qwen 3.8 model. You can try the smaller versions, but even more capability will get left on the table that way.

    Mixed · Sep 2026 · in “Astra and Fable still hack on simple variants of alignment evals from 2025”

40. I was able to see better or comparable results from Qwen3.5-122B-A10B as I got from Qwen3.5-27B, however at a slower generation speed

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