Top all-time · #1140
Qwen
Company blog · qwenlm.github.io/blogaiprogramming
Qwen is the company blog at qwenlm.github.io, writing about ai and programming. Since 2024, 59 different Hacker News readers have linked it 76 times, most often “QwQ: Alibaba's O1-like reasoning LLM”. It is still writing: the last post was on 2025-09-22, and you can follow it by feed.
Qwickly forging AGI, enhancing intelligence.
Most-linked posts?
- QwQ: Alibaba's O1-like reasoning LLM4 people · 4 mentions · first linked 2024-11-27
- Qwen2-VL Can understand video content lasting up to 20 minutes or more4 people · 4 mentions · first linked 2024-08-29
- Qwen2.5-VL4 people · 4 mentions · first linked 2025-01-27
- Qwen2.5-Max: Exploring the intelligence of large-scale MoE model4 people · 4 mentions · first linked 2025-01-28
- Qwen3: Think deeper, act faster3 people · 3 mentions · first linked 2025-04-28
- <think> </think> QwQ-Max-Preview3 people · 3 mentions · first linked 2025-02-24
- Qwen QVQ: First open source 'thinking' vision model.3 people · 3 mentions · first linked 2024-12-24
- Qwen1.5-Moe: Matching 7B Model Performance with 1/3 Activated Parameters3 people · 3 mentions · first linked 2024-03-28
- Qwen2.5-1M: Deploy your own Qwen with context length up to 1M tokens2 people · 2 mentions · first linked 2025-01-26
- Qwen2.5: A Party of Foundation Models2 people · 2 mentions · first linked 2024-09-18
What Hacker News says?
> Unfortunately Gemini really seems to struggle on this, and no matter how we tried prompting it, it would generate wildly inaccurate bounding boxes Qwen2.5 VL was trained on a special HTML format for doing OCR with bounding boxes. [1] The resulting boxes aren't quite as accurate as something like Textract/Surya, but I've found they're much more accurate than Gemini or any other LLM. [1] https://qwenlm.github.io/blo…
Comment linking /blog/qwen2.5-vl/
on “Ingesting PDFs and why Gemini 2.0 changes everything”Submitted Qwen3: Think deeper, act faster as a story · 869 points
HN discussionJust add "/no_think" in your prompt. https://qwenlm.github.io/blog/qwen3/#advanced-usages
Comment linking /blog/qwen3/
on “Qwen3: Think deeper, act faster”Submitted Qwen3-Coder: Agentic coding in the world as a story · 765 points
HN discussionI'm not sure to which extent this opinion is accurately informed. It is well known that nobody trains on 1M token-long content. It wouldn't work anyway as the dependencies are too far fetched and you end up with vanishing gradients. RoPE (Rotary Positional Embeddings, think modulo or periodic arithmetics) scaling is key, whereby the model is trained on 16k tokens long content, and then scaled up to 100k+ [0]. Qwen 1…
Comment linking /blog/qwen2.5-turbo/
on “GPT-4.1 in the API”Submitted Qwen-Image: Crafting with native text rendering as a story · 544 points
HN discussionOutside of Google Deepmind open sourcing the code and weights of AlphaFold, I don't think they've released any of their GenAI stuff (Imagen, Gemini, Flash 2.5, etc). The best multimodal models that you can run locally right now are probably Qwen-Edit 20b, and Kontext.Dev. https://qwenlm.github.io/blog/qwen-image-edit https://bfl.ai/blog/flux-1-kontext-dev
Comment linking /blog/qwen-image-edit
on “Nano Banana image examples”Submitted Qwen2.5-VL-32B: Smarter and Lighter as a story · 544 points
HN discussionI've been following the state of things, but I'm not sure which ones are the best. There's Meta's CodeLlama[1], Mistral's Codestral[2], DeepSeek AI's DeepSeek-Coder-V2-Instruct[3], CodeGemma[4], Alibaba's CodeQwen[5], and Microsoft's WizardCoder[6]. I'm pretty sure CodeLlama is out of date now. I've heard DeepSeek LLMs are good and DeepSeek-Coder-V2-Instruct was released recently. With the good reputation and its ma…
Comment linking /blog/codeqwen1.5/
on “Codestral Mamba”Submitted QwQ-32B: Embracing the Power of Reinforcement Learning as a story · 480 points
HN discussionhttps://qwenlm.github.io/blog/qwen3/ Go look at the benchmark numbers of qwen3-4B if you think these are unrealistic.
Comment linking /blog/qwen3/
on “Xiaomi MiMo Reasoning Model”Submitted QwQ: Alibaba's O1-like reasoning LLM as a story · 438 points
HN discussionSee QwQ-Max-Preview: https://qwenlm.github.io/blog/qwq-max-preview/
Comment linking /blog/qwq-max-preview/
on “QwQ-32B: Embracing the Power of Reinforcement Learning”Submitted Qwen2.5-1M: Deploy your own Qwen with context length up to 1M tokens as a story · 307 points
HN discussionWasn't this release in Nov 2024 as a "preview" with similarly impressive performance? https://qwenlm.github.io/blog/qwq-32b-preview/
Comment linking /blog/qwq-32b-preview/
on “QwQ-32B: Embracing the Power of Reinforcement Learning”Submitted Qwen2 LLM Released as a story · 261 points
HN discussionThe PR confuse a but 32k/64k and 131k if paid API. Also this model https://huggingface.co/Qwen/Qwen3-235B-A22B Is native 32k. So the 64k and 131k use ROPE that is not the best for effective context. While https://qwenlm.github.io/blog/qwen3-coder/ it's 256k native https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct .
Comment linking /blog/qwen3-coder/
on “Cerebras launches Qwen3-235B, achieving 1.5k tokens per second”Submitted Qwen VLo: From “Understanding” the World to “Depicting” It as a story · 223 points
HN discussionThis API only model with a 1M context window was released back in Nov. Just for some historical context. https://qwenlm.github.io/blog/qwen2.5-turbo/
Comment linking /blog/qwen2.5-turbo/
on “Qwen2.5-1M: Deploy your own Qwen with context length up to 1M tokens”Submitted Qwen2.5: A Party of Foundation Models as a story · 168 points
HN discussionThe 32b sounds like it has some useful small tweakers. Tweaks to make output more human friendly, better mathematical reasoning, better fine-grained understanding. https://qwenlm.github.io/blog/qwen2.5-vl-32b/ https://news.ycombinator.com/item?id=43464068 Qwen2.5-VL-72b was released two months ago (to little fanfare in submissions, i think, but some very enthusiastic comments such as rabid enthusiasm for handwriting…
Comment linking /blog/qwen2.5-vl/
on “Show HN: Qwen-2.5-32B is now the best open source OCR model”Submitted Qwen2-Math as a story · 128 points
HN discussionThey are actually very good, cf: https://qwenlm.github.io/blog/qwen-vl/ and it's possible to unalign them (like Dolphin-Yi).
Comment linking /blog/qwen-vl/
on “Mistral Remove "Committing to open models" from their website”Submitted QVQ-Max: Think with Evidence as a story · 121 points
HN discussionI'd also like to point out that they omit Qwen2.5 14B from the benchmark because it doesn't fit their narrative(MMLU Pro score of 63.7[0]). This kind of listing-only-models-you-beat feels extremely shady to me. [0] https://qwenlm.github.io/blog/qwen2.5/
Comment linking /blog/qwen2.5/
on “Liquid Foundation Models: Our First Series of Generative AI Models”Submitted Qwen2.5-Max: Exploring the intelligence of large-scale MoE model as a story · 118 points
HN discussionI'd really like to play with Qwen2.5-VL at some point, perhaps for reading data-sheets for microchips. Nicely for some applications, it's also very good at reporting position of what it finds, which many ML tools are pretty mediocre at. https://qwenlm.github.io/blog/qwen2.5-vl/ Not really this application, but QvQ for visual reasoning is also impressive. https://qwenlm.github.io/blog/qvq-72b-preview/ Meta has used Q…
Comment linking /blog/qvq-72b-preview/
on “Run structured extraction on documents/images locally with Ollama and Pydantic”Submitted Extending the context length to 1M tokens as a story · 116 points
HN discussionThis might be what you are asking for... https://qwenlm.github.io/blog/qwen2.5-coder/ Ctrl F - Code Reasoning:
Comment linking /blog/qwen2.5-coder/
on “Qwen2.5: A Party of Foundation Models”Submitted Qwen1.5-110B as a story · 114 points
HN discussion
