Qwen
Alibaba's family of open-weight large language models, available in many sizes and via hosted chat.
HN commenters like Qwen for running well locally on consumer GPUs and Macs123, for strong quality relative to its size456, and as a cheaper alternative to Anthropic and OpenAI.789 The most common complaint is unreliable agentic coding, such as getting stuck in thought loops or not writing files101112, and it still trails frontier models like Opus.1314
Why people recommend it
Local hardware · 3
Runs usefully on modest local setups, from 16GB VRAM cards to M-series Macs.
“A PC with an nvidia card with 16gb vram works just fine for Qwen MoE models, and these have worked great as a daily driver for me.”
“I use Owen on my MacBook m1 (16gb) daily, running inside Ollama. Works well.”
Show 1 more
“A 24gb vram, 5 year old titan run 27b at 30t/s and the results are good.”
Quality for size · 3
Commenters call it the best or state of the art in its size range.
“It can be considered SOTA within is size category. Very useful for many things.”
“Qwen models are generally the best if you want local models in the 100b max size.”
Show 1 more
“The Qwen models seem to be the best currently in my experience.”
Coding and review · 3
Used for refactoring, code review, and support tasks around coding.
“I use local models like qwen to do refactoring; the code and concept already exist, they need refining.”
“Qwen (at least for the things I use it for) edges just ahead”
Show 1 more
“I use local Qwen to babysit these sessions, to make initial investigation, a repro case, and file a ticket.”
Low cost · 3
Some commenters dropped OpenAI and Anthropic for Qwen and other open models to save money.
“I also use QWEN and GLM for many of the api requiring projects”
“I recommend DeepSeek, GLM/Z-AI or Qwen.”
Show 1 more
“the rest Qwen/Kimi/Deepseek can do”
Privacy · 3
Running it locally keeps sensitive or personal data in-house.
“I use the 35B MoE and 27B dense models locally and most of the time I don’t need to reach out to Claude.”
“expose this data safely, privately, and securely to something like Qwen or Gemma for processing”
Show 1 more
“We do AI Invoice OCR - PDF -> Image -> Excel. Works much better than other solutions”
Range of sizes · 3
Comes in a wide range of model sizes, including very small ones.
“one thing I love with Qwen is it comes a much larger range of sizes so I can experiment which extremely small llms.”
“God I love the Qwen team. Easily the most diverse set of models from all the Chinese labs.”
Show 1 more
“It's my favorite model family to interact with”
Caveats and dissent
Agentic coding · 3
Can loop, write broken code, or fail to write files in agentic coding setups.
“Qwen in particular seems to work at first, then often gets stuck in thought loops when used for actual work.”
“it can do the work just fine - like creating a script - but it just outputs it, rather than placing it into a file.”
Show 1 more
“all the models seem to love writing uncompileable code, and ending up in fearsome knots over it”
Behind frontier · 3
Commenters say it does not match top proprietary models on harder real-world work.
“Qwen models are impressively good for what they are, are "good enough" for plenty tasks, can be ran locally on decently priced hardware”
“even if you can easily run ds or qwen locally to get very similar results”
Show 1 more
“I ran the same evals against a Qwen36 and it clearly lost to both of them”
Needs tuning · 3
Some users had to adjust temperature or prompt format to get good results.
“Qwen on temperatures > 0.5 is unusable for coding and I found sweet spot around 0.32 for coding.”
“I use Owen on my MacBook m1 (16gb) daily, running inside Ollama. Works well.”
Show 1 more
“First, I used qwen without any ChatML-like syntax and it continued speaking and speaking”
Local speed · 3
Some found local performance slow, especially with long contexts.
“Qwen and Deepseek V4 Flash have the same issue”
“the perf of LLM models like Qwen are not ideal for local execution”
Show 1 more
“I use Owen on my MacBook m1 (16gb) daily, running inside Ollama. Works well.”
Tooling quirks · 1
One user reports Qwen does not work well with oMLX.
“Not sure what it is about oMLX + qwen but they don't seem to play well together.”
Multimodal transcription · 1
One user got subpar results for multi-speaker transcription.
“Qwen and other open multimodal models gives subpar results.”
Often compared with
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? 6 mentions
- Running local models is good now 5 mentions
- Qwen3.8-27B 3 mentions
- How to Setup a Local Coding Agent on macOS 3 mentions
- Extracting Steering Vectors from J space 2 mentions
- Anthropic's best AI model struggles to attract users as cheaper tools thrive 2 mentions
Data: qwen.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 47
“I use granite and qwen together with a manually curated set of books, articles, etc. to get work done while disconnected.”
“Preferred Qwen.”
“even if you can easily run ds or qwen locally to get very similar results”
“all the models seem to love writing uncompileable code, and ending up in fearsome knots over it”
“Even the dated qwen models from almost a year ago 1/10th the size are a better user experience”
“still be able to chit-chat with Qwen model”
“(well it's qwen with toolcalls)”
“the perf of LLM models like Qwen are not ideal for local execution”
“a BF16 (unquantized) Qwen beats a Q8 of double its size for decisions”
“It's my favorite model family to interact with”
“on my setup I found it a bit slow for my use case, and the benefits weren't strong enough to outweigh the slowdown”
“I also use QWEN and GLM for many of the api requiring projects”
“I use this coding agent with qwen 2.5b”
“First, I used qwen without any ChatML-like syntax and it continued speaking and speaking”
“God I love the Qwen team. Easily the most diverse set of models from all the Chinese labs.”
“it can do the work just fine - like creating a script - but it just outputs it, rather than placing it into a file.”
“I ran the same evals against a Qwen36 and it clearly lost to both of them”
“one thing I love with Qwen is it comes a much larger range of sizes so I can experiment which extremely small llms.”
“It runs locally by default with Qwen LLMs as the main engine.”
“Qwen (at least for the things I use it for) edges just ahead”
“using Qwen and Gemma models”
“I don't think any Mistral model can match an open 30B-sized Qwen from a year ago”
“I self computed a j lens for the 27b version of the qwen model.”
“I use local Qwen to babysit these sessions, to make initial investigation, a repro case, and file a ticket.”
“I use qwen at my house, it’s pretty great.”
“I use Qwen/GPT-OSS on my Mac Studio via llama.cpp server.”
“send simpler work from Codex (sol xhigh) to cheaper models (Luna, Qwen)”
“I use Qwen/Claude for coding and Gemini/Gemma to draft text intended for a human.”
“the rest Qwen/Kimi/Deepseek can do”
“I'm glad that Kimi K3 and GPT 5.6 Sol and the latest GLM and Qwen and...I love that this has forced Anthropic to change plans”
“I have been using open source LLMs locally for the past 5 months like Qwen, Llama, DeepSeek and others”
“Qwen running locally on a 64GB M1 Max Macbook Pro to help save my ever-quickly shrinking Codex credits”
“The Qwen models are decent, but they don't come close to the full Claude experience I've come to expect.”
“it's still a problem with the Qwen and Deepseek models I use now”
“I use Qwen mostly for vision in my personal assistant.”
“I use the Qwens as a vision model for my DeepSeek V4 Flashes to handle.”
“actually prefer the open weight models in most day to day usage”
“Qwen models are impressively good for what they are, are "good enough" for plenty tasks, can be ran locally on decently priced hardware”
“put a few qwen models for vision and embedding purposes on my original DGX Spark”
“Codex CLI, Claude Code, Qwen”
“It can be considered SOTA within is size category. Very useful for many things.”
“A PC with an nvidia card with 16gb vram works just fine for Qwen MoE models, and these have worked great as a daily driver for me.”
“Qwen in particular seems to work at first, then often gets stuck in thought loops when used for actual work.”
“I recommend DeepSeek, GLM/Z-AI or Qwen.”
“Qwen models are generally the best if you want local models in the 100b max size.”
“A 24gb vram, 5 year old titan run 27b at 30t/s and the results are good.”
“I use Owen on my MacBook m1 (16gb) daily, running inside Ollama. Works well.”