{"slug": "qwen", "name": "Qwen", "category": "ai", "category_name": "AI models and assistants", "description": "Alibaba's family of open-weight large language models, available in many sizes and via hosted chat.", "recommenders": 41, "commenters": 96, "warners": 7, "score": 42.805, "rising": 8.49, "first": "2025-11-14", "latest": "2026-10-28", "why": "HN commenters like Qwen for running well locally on consumer GPUs and Macs, for strong quality relative to its size, and as a cheaper alternative to Anthropic and OpenAI.", "why_is_quote": false, "has_verdict": true, "sentiment": {"recommend": 42, "neutral": 44, "compare": 5, "warn": 8}, "spark": [0, 1, 1, 0, 0, 1, 4, 11, 7, 6, 6, 5], "listed": true, "loved": ["Local hardware", "Quality for size"], "caveat": "Agentic coding", "short_quote": "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.", "icon": true, "domain": null, "github_repo": null, "official_url": "https://qwen.ai", "pricing": "open source", "aliases": ["Qwen"], "verdict": [{"text": "HN commenters like Qwen for running well locally on consumer GPUs and Macs", "cites": [{"id": 780156032, "item_id": 48759752, "url": "https://news.ycombinator.com/item?id=48759752", "by": "jadbox", "date": "2026-07-02", "stance": "recommend", "quote": "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.", "excerpt": "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.", "context": "", "use_case": "local daily driver model", "switched_from": "", "story": "Kimi K2.7 Code is generally available in GitHub Copilot"}, {"id": 776725328, "item_id": 48545333, "url": "https://news.ycombinator.com/item?id=48545333", "by": "major505", "date": "2026-06-15", "stance": "recommend", "quote": "I use Owen on my MacBook m1 (16gb) daily, running inside Ollama. Works well.", "excerpt": "Yes. I use Owen on my MacBook m1 (16gb) daily, running inside Ollama. Works well. Is not particularly fast, and I need to create a custom imagem that sets the temperature of the model to zero starting, so I don't get over creative with its bullshit, but it works reasonable week.", "context": "", "use_case": "local daily coding", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}, {"id": 777018560, "item_id": 48563660, "url": "https://news.ycombinator.com/item?id=48563660", "by": "qudat", "date": "2026-06-16", "stance": "recommend", "quote": "A 24gb vram, 5 year old titan run 27b at 30t/s and the results are good.", "excerpt": "I feel like the claims come from wildly different personas and use cases. A 24gb vram, 5 year old titan run 27b at 30t/s and the results are good. I use sonnet and opus at my day job and they are more capable but I can still get the same out of qwen, I just need to be mindful of ctx", "context": "uses Sonnet and Opus at day job", "use_case": "local inference", "switched_from": "", "story": "Running local models is good now"}]}, {"text": ", for strong quality relative to its size", "cites": [{"id": 780461297, "item_id": 48778831, "url": "https://news.ycombinator.com/item?id=48778831", "by": "verdverm", "date": "2026-07-03", "stance": "recommend", "quote": "It can be considered SOTA within is size category. Very useful for many things.", "excerpt": "It can be considered SOTA within is size category. Very useful for many things. You still want access to big models, I recommend OpenCode Go if you want to stay with open models.", "context": "", "use_case": "local LLM", "switched_from": "", "story": "Jamesob's guide to running SOTA LLMs locally"}, {"id": 777377792, "item_id": 48586112, "url": "https://news.ycombinator.com/item?id=48586112", "by": "Imustaskforhelp", "date": "2026-06-18", "stance": "recommend", "quote": "Qwen models are generally the best if you want local models in the 100b max size.", "excerpt": "Qwen models are generally the best if you want local models in the 100b max size. I am not sure but I recommend looking at the size of deepseek flash as well to see if it can work for you, perhaps GLM 5 Air model is also around this size. and if you want something at the larger models scale, then you have glm 5.2, kimi k2.7,mimo v2.5 pro and deepseek v4", "context": "", "use_case": "local models", "switched_from": "", "story": "Ask HN: Whats the best and small open source model?"}, {"id": 793835361, "item_id": 49614710, "url": "https://news.ycombinator.com/item?id=49614710", "by": "moffkalast", "date": "2026-09-08", "stance": "recommend", "quote": "I don't think any Mistral model can match an open 30B-sized Qwen from a year ago", "excerpt": "I don't think any Mistral model can match an open 30B-sized Qwen from a year ago, so right now they're not really a competitive alternative to anything at all. Except Voxtral perhaps, but that's very niche.", "context": "", "use_case": "", "switched_from": "", "story": "Mistral raises \u20ac3B"}]}, {"text": ", and as a cheaper alternative to Anthropic and OpenAI.", "cites": [{"id": 798349120, "item_id": 49896820, "url": "https://news.ycombinator.com/item?id=49896820", "by": "system2", "date": "2026-09-29", "stance": "recommend", "quote": "I also use QWEN and GLM for many of the api requiring projects", "excerpt": "Chinese model pressure. Many of my SWE friends switched to Chinese models. I also use QWEN and GLM for many of the api requiring projects and dropped OpenAI and Anthropic. The only reason was the cost. EDIT: I love getting downvoted by openai and anthropic employees or their bots.", "context": "", "use_case": "API-based projects", "switched_from": "", "story": "GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price"}, {"id": 778266354, "item_id": 48641647, "url": "https://news.ycombinator.com/item?id=48641647", "by": "yenihesap", "date": "2026-06-23", "stance": "recommend", "quote": "I recommend DeepSeek, GLM/Z-AI or Qwen.", "excerpt": "Just use one of the alternatives, there are so many. I recommend DeepSeek, GLM/Z-AI or Qwen. They are cheaper and offer competitive performance to Anthropic.", "context": "", "use_case": "coding assistant alternative", "switched_from": "", "story": "Ask HN: Anthropic banned me from using Claude Code and I don't know what to do"}, {"id": 790890785, "item_id": 49430674, "url": "https://news.ycombinator.com/item?id=49430674", "by": "ciefa", "date": "2026-08-25", "stance": "recommend", "quote": "the rest Qwen/Kimi/Deepseek can do", "excerpt": "I switched to GPT 5.6 Sol for the hardest tasks, the rest Qwen/Kimi/Deepseek can do. So far not looking back to Anthropic. They lost me.", "context": "", "use_case": "routine tasks", "switched_from": "", "story": "Anthropic's best AI model struggles to attract users as cheaper tools thrive"}]}, {"text": " The most common complaint is unreliable agentic coding, such as getting stuck in thought loops or not writing files", "cites": [{"id": 779566403, "item_id": 48722900, "url": "https://news.ycombinator.com/item?id=48722900", "by": "blopker", "date": "2026-06-29", "stance": "warn", "quote": "Qwen in particular seems to work at first, then often gets stuck in thought loops when used for actual work.", "excerpt": "I've been working with local models for the past year. There's so many possibilities, but I don't think coding is one. Coding requires so many layers beyond inference; I spent so much time trying to replicate what Claude Code does end to end locally. Understanding all the layers and keeping up with the advancements feels like a slog. Even this article messes up and misunderstands what some of the settings are doing. ", "context": "", "use_case": "local coding", "switched_from": "", "story": "Qwen 3.6 27B is the sweet spot for local development"}, {"id": 796128082, "item_id": 49758005, "url": "https://news.ycombinator.com/item?id=49758005", "by": "HoldOnAMinute", "date": "2026-09-18", "stance": "warn", "quote": "it can do the work just fine - like creating a script - but it just outputs it, rather than placing it into a file.", "excerpt": "No - Here is what I mean. When I use Claude Code locally with Ollama/Quen, it can do the work just fine - like creating a script - but it just outputs it, rather than placing it into a file. So far I can't convince it to actually WRITE to a file. With Claude Code, it just works. Recursing around the file system, it just does it. I can't get that working locally.", "context": "", "use_case": "local coding model", "switched_from": "", "story": "Ask HN: What happens when Claude Code goes away?"}, {"id": 802356163, "item_id": 50147260, "url": "https://news.ycombinator.com/item?id=50147260", "by": "Perenti", "date": "2026-10-19", "stance": "warn", "quote": "all the models seem to love writing uncompileable code, and ending up in fearsome knots over it", "excerpt": "I love Free Pascal. I've tried using claude/codex/qwen to write in it, but all the models seem to love writing uncompileable code, and ending up in fearsome knots over it. It's my number one \"knock something together that just works\" language for humans though.", "context": "", "use_case": "writing Free Pascal code", "switched_from": "", "story": "Free Pascal compiler now supports dynamic packages"}]}, {"text": ", and it still trails frontier models like Opus.", "cites": [{"id": 788800640, "item_id": 49300040, "url": "https://news.ycombinator.com/item?id=49300040", "by": "NitpickLawyer", "date": "2026-08-14", "stance": "recommend", "quote": "Qwen models are impressively good for what they are, are \"good enough\" for plenty tasks, can be ran locally on decently priced hardware", "excerpt": "> Beats Opus 4.7 Max I'm a huge open model fan, and have used them since forever, even have daily drivers for on-prem dev, but no. They do not beat opus on real-world usage. Qwen models are impressively good for what they are, are \"good enough\" for plenty tasks, can be ran locally on decently priced hardware, and so on. They certainly have their uses, and the field in general has advanced faster than my early expecta", "context": "open model fan with on-prem dev", "use_case": "local on-prem dev", "switched_from": "", "story": "Qwen3.8-27B"}, {"id": 802560322, "item_id": 50160020, "url": "https://news.ycombinator.com/item?id=50160020", "by": "mlcruz", "date": "2026-10-20", "stance": "neutral", "quote": "even if you can easily run ds or qwen locally to get very similar results", "excerpt": "I think we keep getting used to intelligence improvements and the goal posts are always moving forward. I remember that close to an year ago i felt the same way about opus 4.5. And even if you can easily run ds or qwen locally to get very similar results to that, we moved to 5.5 being the new target. Deepseek can do 90% of the job but a claude reviwer/spec writer is a brutal improvement.", "context": "", "use_case": "running locally", "switched_from": "", "story": "MiniMax M3.1 Flash"}]}], "praise": [{"label": "Local hardware", "summary": "Runs usefully on modest local setups, from 16GB VRAM cards to M-series Macs.", "quotes": [{"id": 780156032, "item_id": 48759752, "url": "https://news.ycombinator.com/item?id=48759752", "by": "jadbox", "date": "2026-07-02", "stance": "recommend", "quote": "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.", "excerpt": "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.", "context": "", "use_case": "local daily driver model", "switched_from": "", "story": "Kimi K2.7 Code is generally available in GitHub Copilot"}, {"id": 776725328, "item_id": 48545333, "url": "https://news.ycombinator.com/item?id=48545333", "by": "major505", "date": "2026-06-15", "stance": "recommend", "quote": "I use Owen on my MacBook m1 (16gb) daily, running inside Ollama. Works well.", "excerpt": "Yes. I use Owen on my MacBook m1 (16gb) daily, running inside Ollama. Works well. Is not particularly fast, and I need to create a custom imagem that sets the temperature of the model to zero starting, so I don't get over creative with its bullshit, but it works reasonable week.", "context": "", "use_case": "local daily coding", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}, {"id": 777018560, "item_id": 48563660, "url": "https://news.ycombinator.com/item?id=48563660", "by": "qudat", "date": "2026-06-16", "stance": "recommend", "quote": "A 24gb vram, 5 year old titan run 27b at 30t/s and the results are good.", "excerpt": "I feel like the claims come from wildly different personas and use cases. A 24gb vram, 5 year old titan run 27b at 30t/s and the results are good. I use sonnet and opus at my day job and they are more capable but I can still get the same out of qwen, I just need to be mindful of ctx", "context": "uses Sonnet and Opus at day job", "use_case": "local inference", "switched_from": "", "story": "Running local models is good now"}]}, {"label": "Quality for size", "summary": "Commenters call it the best or state of the art in its size range.", "quotes": [{"id": 780461297, "item_id": 48778831, "url": "https://news.ycombinator.com/item?id=48778831", "by": "verdverm", "date": "2026-07-03", "stance": "recommend", "quote": "It can be considered SOTA within is size category. Very useful for many things.", "excerpt": "It can be considered SOTA within is size category. Very useful for many things. You still want access to big models, I recommend OpenCode Go if you want to stay with open models.", "context": "", "use_case": "local LLM", "switched_from": "", "story": "Jamesob's guide to running SOTA LLMs locally"}, {"id": 777377792, "item_id": 48586112, "url": "https://news.ycombinator.com/item?id=48586112", "by": "Imustaskforhelp", "date": "2026-06-18", "stance": "recommend", "quote": "Qwen models are generally the best if you want local models in the 100b max size.", "excerpt": "Qwen models are generally the best if you want local models in the 100b max size. I am not sure but I recommend looking at the size of deepseek flash as well to see if it can work for you, perhaps GLM 5 Air model is also around this size. and if you want something at the larger models scale, then you have glm 5.2, kimi k2.7,mimo v2.5 pro and deepseek v4", "context": "", "use_case": "local models", "switched_from": "", "story": "Ask HN: Whats the best and small open source model?"}, {"id": 773446048, "item_id": 48340378, "url": "https://news.ycombinator.com/item?id=48340378", "by": "nulledy", "date": "2026-05-30", "stance": "recommend", "quote": "The Qwen models seem to be the best currently in my experience.", "excerpt": "The Qwen models seem to be the best currently in my experience. But if you're thinking about buying new hardware just for local inference: don't. You'll spend more upfront for rapidly out of date and depreciating hardware while running inferior models compared to the latest from Anthropic and OpenAI. Unless your requirement is prompt privacy, or you have idle hardware sitting around, the ROI just isn't there compared", "context": "", "use_case": "local inference", "switched_from": "", "story": "Ask HN: If I cancel Codex today whats the next best local inference agent?"}]}, {"label": "Coding and review", "summary": "Used for refactoring, code review, and support tasks around coding.", "quotes": [{"id": 785644448, "item_id": 49102778, "url": "https://news.ycombinator.com/item?id=49102778", "by": "serf", "date": "2026-07-29", "stance": "recommend", "quote": "I use local models like qwen to do refactoring; the code and concept already exist, they need refining.", "excerpt": "I use local models like qwen to do refactoring; the code and concept already exist, they need refining. I use sota models for concept, implementation plans, and initial deployment. I find that local models can perform at a very high level if given a very strict environment and the already bounded rules of an existing code base. aye aye aye.", "context": "", "use_case": "local code refactoring", "switched_from": "", "story": "Self-hosting Kimi K3: 20% more hardware cost, 20% better task resolution"}, {"id": 795045376, "item_id": 49690336, "url": "https://news.ycombinator.com/item?id=49690336", "by": "noir_lord", "date": "2026-09-14", "stance": "recommend", "quote": "Qwen (at least for the things I use it for) edges just ahead", "excerpt": "I did, Muse came close but what impressed me about Qwen is it\u2019ll push back if it thinks it\u2019s right even when it isn\u2019t, I can work with that, Muse tended to flip between states too easily/too much. It is a good model but Qwen (at least for the things I use it for) edges just ahead, it seems much better at the \u201crip this apart, suggest improvements, touch nothing\u201d use case where I can use it as a second set of eyes, I d", "context": "runs local models on a 7900XTX", "use_case": "local code review", "switched_from": "", "story": "Astra and Fable still hack on simple variants of alignment evals from 2025"}, {"id": 792444849, "item_id": 49527803, "url": "https://news.ycombinator.com/item?id=49527803", "by": "aka-rider", "date": "2026-09-01", "stance": "recommend", "quote": "I use local Qwen to babysit these sessions, to make initial investigation, a repro case, and file a ticket.", "excerpt": "All very good questions. Agents are actively destroy QA gates in many ways, usually by cheating (\"the test is buggy, not my changes\" \u2014 changes the test), or just rot QA slowly by writing buggy overcomplicated tests What works for me 10/10 is fuzzing and my own constant usage. For this project specifically (text editor), I asked LLM to create human-like fuzzing session, it sends keystrokes like: \"the user is searching", "context": "", "use_case": "babysitting fuzzing sessions", "switched_from": "", "story": "I used Fable to rewrite 65kLoC of Go in Rust. It cost $400"}]}, {"label": "Low cost", "summary": "Some commenters dropped OpenAI and Anthropic for Qwen and other open models to save money.", "quotes": [{"id": 798349120, "item_id": 49896820, "url": "https://news.ycombinator.com/item?id=49896820", "by": "system2", "date": "2026-09-29", "stance": "recommend", "quote": "I also use QWEN and GLM for many of the api requiring projects", "excerpt": "Chinese model pressure. Many of my SWE friends switched to Chinese models. I also use QWEN and GLM for many of the api requiring projects and dropped OpenAI and Anthropic. The only reason was the cost. EDIT: I love getting downvoted by openai and anthropic employees or their bots.", "context": "", "use_case": "API-based projects", "switched_from": "", "story": "GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price"}, {"id": 778266354, "item_id": 48641647, "url": "https://news.ycombinator.com/item?id=48641647", "by": "yenihesap", "date": "2026-06-23", "stance": "recommend", "quote": "I recommend DeepSeek, GLM/Z-AI or Qwen.", "excerpt": "Just use one of the alternatives, there are so many. I recommend DeepSeek, GLM/Z-AI or Qwen. They are cheaper and offer competitive performance to Anthropic.", "context": "", "use_case": "coding assistant alternative", "switched_from": "", "story": "Ask HN: Anthropic banned me from using Claude Code and I don't know what to do"}, {"id": 790890785, "item_id": 49430674, "url": "https://news.ycombinator.com/item?id=49430674", "by": "ciefa", "date": "2026-08-25", "stance": "recommend", "quote": "the rest Qwen/Kimi/Deepseek can do", "excerpt": "I switched to GPT 5.6 Sol for the hardest tasks, the rest Qwen/Kimi/Deepseek can do. So far not looking back to Anthropic. They lost me.", "context": "", "use_case": "routine tasks", "switched_from": "", "story": "Anthropic's best AI model struggles to attract users as cheaper tools thrive"}]}, {"label": "Privacy", "summary": "Running it locally keeps sensitive or personal data in-house.", "quotes": [{"id": 783461216, "item_id": 48966326, "url": "https://news.ycombinator.com/item?id=48966326", "by": "nsbk", "date": "2026-07-19", "stance": "recommend", "quote": "I use the 35B MoE and 27B dense models locally and most of the time I don\u2019t need to reach out to Claude.", "excerpt": "Bring it on! Hoping that they release smaller sizes of Qwen3.8. I use the 35B MoE and 27B dense models locally and most of the time I don\u2019t need to reach out to Claude. Extremely useful specially when requests include sensitive and/or personal data", "context": "", "use_case": "local LLM for private data", "switched_from": "", "story": "Qwen3.8 is launching and going open-weight soon"}, {"id": 777341377, "item_id": 48583836, "url": "https://news.ycombinator.com/item?id=48583836", "by": "stego-tech", "date": "2026-06-18", "stance": "recommend", "quote": "expose this data safely, privately, and securely to something like Qwen or Gemma for processing", "excerpt": "I still believe that the strength of AI is when it can be applied locally in a secure and private manner, rather than yet another cloud-based service you must pay for indefinitely even as it gets progressively worse to satiate the greed of corporate shareholders. ChatGPT and Anthropic will never, ever get me to tie my Health Data to their systems, but I still believe in the capabilities of AI in identifying patterns ", "context": "", "use_case": "local health data analysis", "switched_from": "", "story": "Local Qwen isn't a worse Opus, it's a different tool"}, {"id": 744973632, "item_id": 46560852, "url": "https://news.ycombinator.com/item?id=46560852", "by": "delaminator", "date": "2026-01-09", "stance": "recommend", "quote": "We do AI Invoice OCR - PDF -> Image -> Excel. Works much better than other solutions", "excerpt": "I have a 3090 24Gb Twin Xenon 64Gb RAM sat on a machine in our server room. I do local AI with Qwen, Whisper and another I can't remember right now. These are all QWEN: We do AI Invoice OCR - PDF -> Image -> Excel. Works much better than other solutions because it has invoice context so it looks for particular data to extract and ignores others. Why local? I proved it worked, no need to send our data outside for proc", "context": "runs local AI on a 3090 at work", "use_case": "invoice OCR, ingredient classification", "switched_from": "", "story": "Ask HN: Who's running local AI workstations in 2026?"}]}, {"label": "Range of sizes", "summary": "Comes in a wide range of model sizes, including very small ones.", "quotes": [{"id": 795986592, "item_id": 49749162, "url": "https://news.ycombinator.com/item?id=49749162", "by": "tolugenius", "date": "2026-09-18", "stance": "recommend", "quote": "one thing I love with Qwen is it comes a much larger range of sizes so I can experiment which extremely small llms.", "excerpt": "Curious if or when we'll see the Qwen4 series, one thing I love with Qwen is it comes a much larger range of sizes so I can experiment which extremely small llms.", "context": "", "use_case": "experimenting with small LLMs", "switched_from": "", "story": "Qwen 3.8 Omni Flash"}, {"id": 796414736, "item_id": 49775921, "url": "https://news.ycombinator.com/item?id=49775921", "by": "d2kx", "date": "2026-09-20", "stance": "recommend", "quote": "God I love the Qwen team. Easily the most diverse set of models from all the Chinese labs.", "excerpt": "God I love the Qwen team. Easily the most diverse set of models from all the Chinese labs. Only Gemini/DeepMind comes close.", "context": "", "use_case": "", "switched_from": "", "story": "Qwen Image 2.1"}, {"id": 799001152, "item_id": 49937572, "url": "https://news.ycombinator.com/item?id=49937572", "by": "verdverm", "date": "2026-10-02", "stance": "recommend", "quote": "It's my favorite model family to interact with", "excerpt": "I use all the qwen! It's my favorite model family to interact with, it's prose is the best imo, it makes me laugh from time-to-time (like when it said it would \"crib\" some code from another project, lul) I currently have qwen-flash working on an NES emulator harness so qwen-little can play my first RPG (ff1) (tho I have used all the others I mentioned, happenstance I'm using qwen this iteration/task)", "context": "", "use_case": "coding and agents", "switched_from": "", "story": "One month coding with GLM 5.3 Flash"}]}], "caveats": [{"label": "Agentic coding", "summary": "Can loop, write broken code, or fail to write files in agentic coding setups.", "quotes": [{"id": 779566403, "item_id": 48722900, "url": "https://news.ycombinator.com/item?id=48722900", "by": "blopker", "date": "2026-06-29", "stance": "warn", "quote": "Qwen in particular seems to work at first, then often gets stuck in thought loops when used for actual work.", "excerpt": "I've been working with local models for the past year. There's so many possibilities, but I don't think coding is one. Coding requires so many layers beyond inference; I spent so much time trying to replicate what Claude Code does end to end locally. Understanding all the layers and keeping up with the advancements feels like a slog. Even this article messes up and misunderstands what some of the settings are doing. ", "context": "", "use_case": "local coding", "switched_from": "", "story": "Qwen 3.6 27B is the sweet spot for local development"}, {"id": 796128082, "item_id": 49758005, "url": "https://news.ycombinator.com/item?id=49758005", "by": "HoldOnAMinute", "date": "2026-09-18", "stance": "warn", "quote": "it can do the work just fine - like creating a script - but it just outputs it, rather than placing it into a file.", "excerpt": "No - Here is what I mean. When I use Claude Code locally with Ollama/Quen, it can do the work just fine - like creating a script - but it just outputs it, rather than placing it into a file. So far I can't convince it to actually WRITE to a file. With Claude Code, it just works. Recursing around the file system, it just does it. I can't get that working locally.", "context": "", "use_case": "local coding model", "switched_from": "", "story": "Ask HN: What happens when Claude Code goes away?"}, {"id": 802356163, "item_id": 50147260, "url": "https://news.ycombinator.com/item?id=50147260", "by": "Perenti", "date": "2026-10-19", "stance": "warn", "quote": "all the models seem to love writing uncompileable code, and ending up in fearsome knots over it", "excerpt": "I love Free Pascal. I've tried using claude/codex/qwen to write in it, but all the models seem to love writing uncompileable code, and ending up in fearsome knots over it. It's my number one \"knock something together that just works\" language for humans though.", "context": "", "use_case": "writing Free Pascal code", "switched_from": "", "story": "Free Pascal compiler now supports dynamic packages"}]}, {"label": "Behind frontier", "summary": "Commenters say it does not match top proprietary models on harder real-world work.", "quotes": [{"id": 788800640, "item_id": 49300040, "url": "https://news.ycombinator.com/item?id=49300040", "by": "NitpickLawyer", "date": "2026-08-14", "stance": "recommend", "quote": "Qwen models are impressively good for what they are, are \"good enough\" for plenty tasks, can be ran locally on decently priced hardware", "excerpt": "> Beats Opus 4.7 Max I'm a huge open model fan, and have used them since forever, even have daily drivers for on-prem dev, but no. They do not beat opus on real-world usage. Qwen models are impressively good for what they are, are \"good enough\" for plenty tasks, can be ran locally on decently priced hardware, and so on. They certainly have their uses, and the field in general has advanced faster than my early expecta", "context": "open model fan with on-prem dev", "use_case": "local on-prem dev", "switched_from": "", "story": "Qwen3.8-27B"}, {"id": 802560322, "item_id": 50160020, "url": "https://news.ycombinator.com/item?id=50160020", "by": "mlcruz", "date": "2026-10-20", "stance": "neutral", "quote": "even if you can easily run ds or qwen locally to get very similar results", "excerpt": "I think we keep getting used to intelligence improvements and the goal posts are always moving forward. I remember that close to an year ago i felt the same way about opus 4.5. And even if you can easily run ds or qwen locally to get very similar results to that, we moved to 5.5 being the new target. Deepseek can do 90% of the job but a claude reviwer/spec writer is a brutal improvement.", "context": "", "use_case": "running locally", "switched_from": "", "story": "MiniMax M3.1 Flash"}, {"id": 796093971, "item_id": 49755873, "url": "https://news.ycombinator.com/item?id=49755873", "by": "mmastrac", "date": "2026-09-18", "stance": "compare", "quote": "I ran the same evals against a Qwen36 and it clearly lost to both of them", "excerpt": "If you want to try a _legit_ Jev implementation that matches (at least in my evals), the vLLM patch to turn DiffusionGemma into Jev is available. On my DGX Spark I get very similar latency numbers, and it matches my evals + or - a few points on each test (DG wins some, Jev wins some, both show low confidence when wrong). I ran the same evals against a Qwen36 and it clearly lost to both of them, so you are leaving bot", "context": "", "use_case": "evals", "switched_from": "", "story": "OpenJev"}]}, {"label": "Needs tuning", "summary": "Some users had to adjust temperature or prompt format to get good results.", "quotes": [{"id": 777075521, "item_id": 48567220, "url": "https://news.ycombinator.com/item?id=48567220", "by": "xlii", "date": "2026-06-17", "stance": "recommend", "quote": "Qwen on temperatures > 0.5 is unusable for coding and I found sweet spot around 0.32 for coding.", "excerpt": "> I use a lot of local models and they're still pretty painful to run locally. This really depends on how and what you're using. e.g. I can't suffer through slowness of inference on Macbook but I have gaming rig with quite powerful GPU and I squeeze ~130 t/s on Gemma or ~70t/s on Qwen. Tuning is not optional as well. Qwen on temperatures > 0.5 is unusable for coding and I found sweet spot around 0.32 for coding. Spec", "context": "has gaming rig with powerful GPU", "use_case": "local coding", "switched_from": "", "story": "Running local models is good now"}, {"id": 776725328, "item_id": 48545333, "url": "https://news.ycombinator.com/item?id=48545333", "by": "major505", "date": "2026-06-15", "stance": "recommend", "quote": "I use Owen on my MacBook m1 (16gb) daily, running inside Ollama. Works well.", "excerpt": "Yes. I use Owen on my MacBook m1 (16gb) daily, running inside Ollama. Works well. Is not particularly fast, and I need to create a custom imagem that sets the temperature of the model to zero starting, so I don't get over creative with its bullshit, but it works reasonable week.", "context": "", "use_case": "local daily coding", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}, {"id": 797901168, "item_id": 49868823, "url": "https://news.ycombinator.com/item?id=49868823", "by": "dsjakupov", "date": "2026-09-27", "stance": "neutral", "quote": "First, I used qwen without any ChatML-like syntax and it continued speaking and speaking", "excerpt": "Oh, I had similar case. First, I used qwen without any ChatML-like syntax and it continued speaking and speaking, then I used <|im_start|>/<|im_end|> to control it somehow", "context": "", "use_case": "local LLM", "switched_from": "", "story": "\"As a Language Model\": Chat Template Switches LLM Self-Referential Voice"}]}, {"label": "Local speed", "summary": "Some found local performance slow, especially with long contexts.", "quotes": [{"id": 778859202, "item_id": 48678700, "url": "https://news.ycombinator.com/item?id=48678700", "by": "bigyabai", "date": "2026-06-25", "stance": "warn", "quote": "Qwen and Deepseek V4 Flash have the same issue", "excerpt": "> From GLM 4.7 flash GLM 4.7 Flash is a 30b model that was far behind SOTA at launch, and I know that because I pay for z.ai inference and have run the model locally. Qwen and Deepseek V4 Flash have the same issue, and beg the question; are you really going to process a 64k agentic context at 450tok/s? That's 2+ minutes that you spend waiting for the first token to generate! Of course nobody can sell that as competit", "context": "pays for z.ai inference", "use_case": "local agentic inference", "switched_from": "", "story": "Apple announces significant price increases for MacBooks, iPads, more"}, {"id": 800638737, "item_id": 50039921, "url": "https://news.ycombinator.com/item?id=50039921", "by": "sachaa", "date": "2026-10-11", "stance": "warn", "quote": "the perf of LLM models like Qwen are not ideal for local execution", "excerpt": "This is a great way to start but the perf of LLM models like Qwen are not ideal for local execution. I have adapted Laya (pure decision model) to run in a browser and I am able to get responses under 200ms. Give it a try: https://wexare-ai.github.io/browser-laya/", "context": "", "use_case": "local LLM", "switched_from": "", "story": "Build your own decision model"}, {"id": 776725328, "item_id": 48545333, "url": "https://news.ycombinator.com/item?id=48545333", "by": "major505", "date": "2026-06-15", "stance": "recommend", "quote": "I use Owen on my MacBook m1 (16gb) daily, running inside Ollama. Works well.", "excerpt": "Yes. I use Owen on my MacBook m1 (16gb) daily, running inside Ollama. Works well. Is not particularly fast, and I need to create a custom imagem that sets the temperature of the model to zero starting, so I don't get over creative with its bullshit, but it works reasonable week.", "context": "", "use_case": "local daily coding", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}]}, {"label": "Tooling quirks", "summary": "One user reports Qwen does not work well with oMLX.", "quotes": [{"id": 776157009, "item_id": 48509813, "url": "https://news.ycombinator.com/item?id=48509813", "by": "nozzlegear", "date": "2026-06-12", "stance": "warn", "quote": "Not sure what it is about oMLX + qwen but they don't seem to play well together.", "excerpt": "I use oMLX and also have this issue with qwen. Not sure what it is about oMLX + qwen but they don't seem to play well together.", "context": "", "use_case": "local model with tool calls", "switched_from": "", "story": "Show HN: Script to bulk delete Claude chats from the web UI"}]}, {"label": "Multimodal transcription", "summary": "One user got subpar results for multi-speaker transcription.", "quotes": [{"id": 747879314, "item_id": 46742457, "url": "https://news.ycombinator.com/item?id=46742457", "by": "dvfjsdhgfv", "date": "2026-01-24", "stance": "warn", "quote": "Qwen and other open multimodal models gives subpar results.", "excerpt": "I spent a few days on similar scenario without much success (scenario where one person speaks and then their speech is translated, and I want juts the original or both). An API call to GPT4o works quite well (it basically handles both transcription and diarization), but I wanted a local model. Whisper is really good for 1 person speaking. With more people you get repetitions. Qwen and other open multimodal models giv", "context": "", "use_case": "multi-speaker transcription", "switched_from": "", "story": "Ask HN: What's the current best local/open speech-to-speech setup?"}]}], "who": [{"by": "fultonn", "context": "works offline at a remote property", "url": "https://news.ycombinator.com/item?id=50268236", "date": "2026-10-28"}, {"by": "noir_lord", "context": "runs local models on a 7900XTX", "url": "https://news.ycombinator.com/item?id=49690336", "date": "2026-09-14"}, {"by": "nater5000", "context": "runs a home server with an RTX 3090 Ti", "url": "https://news.ycombinator.com/item?id=49391156", "date": "2026-08-21"}, {"by": "nozzlegear", "context": "F# developer", "url": "https://news.ycombinator.com/item?id=49361923", "date": "2026-08-19"}, {"by": "arjie", "context": "runs local GPUs", "url": "https://news.ycombinator.com/item?id=49300815", "date": "2026-08-14"}, {"by": "NitpickLawyer", "context": "open model fan with on-prem dev", "url": "https://news.ycombinator.com/item?id=49300040", "date": "2026-08-14"}, {"by": "danielnexaaly", "context": "AI-assisted technical operator", "url": "https://news.ycombinator.com/item?id=49251761", "date": "2026-08-10"}, {"by": "kamranjon", "context": "runs local models on Apple Silicon", "url": "https://news.ycombinator.com/item?id=48677006", "date": "2026-06-25"}, {"by": "trollbridge", "context": "small dev team of 6", "url": "https://news.ycombinator.com/item?id=48648590", "date": "2026-06-23"}, {"by": "xlii", "context": "has gaming rig with powerful GPU", "url": "https://news.ycombinator.com/item?id=48567220", "date": "2026-06-17"}, {"by": "qudat", "context": "uses Sonnet and Opus at day job", "url": "https://news.ycombinator.com/item?id=48563660", "date": "2026-06-16"}, {"by": "anubhav200", "context": "runs local models on two machines", "url": "https://news.ycombinator.com/item?id=48557136", "date": "2026-06-16"}], "tags": ["for local inference", "for private data", "for offline work", "for code refactoring", "for document OCR"], "use_cases": ["local llm", "local coding", "local coding model", "local inference", "local model", "coding"], "compared": [{"slug": "gemma", "name": "Gemma", "co": 23, "switched_from": 0, "switched_to": 0}, {"slug": "claude", "name": "Claude", "co": 18, "switched_from": 2, "switched_to": 0}, {"slug": "deepseek", "name": "DeepSeek", "co": 19, "switched_from": 0, "switched_to": 0}, {"slug": "kimi", "name": "Kimi", "co": 14, "switched_from": 1, "switched_to": 0}, {"slug": "claude-opus", "name": "Claude Opus", "co": 14, "switched_from": 0, "switched_to": 0}, {"slug": "glm", "name": "GLM", "co": 12, "switched_from": 0, "switched_to": 0}, {"slug": "opencode", "name": "OpenCode", "co": 11, "switched_from": 0, "switched_to": 0}, {"slug": "ollama", "name": "Ollama", "co": 10, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=48542100", "title": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?", "mentions": 6}, {"url": "https://news.ycombinator.com/item?id=48555993", "title": "Running local models is good now", "mentions": 5}, {"url": "https://news.ycombinator.com/item?id=49299684", "title": "Qwen3.8-27B", "mentions": 3}, {"url": "https://news.ycombinator.com/item?id=48507020", "title": "How to Setup a Local Coding Agent on macOS", "mentions": 3}, {"url": "https://news.ycombinator.com/item?id=49586667", "title": "Extracting Steering Vectors from J space", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=49411102", "title": "Anthropic's best AI model struggles to attract users as cheaper tools thrive", "mentions": 2}], "sentiment_by_month": {"2025-11": {"neutral": 1}, "2025-12": {"recommend": 1}, "2026-01": {"recommend": 1, "warn": 1}, "2026-02": {"neutral": 1}, "2026-03": {"neutral": 1}, "2026-04": {"neutral": 1, "recommend": 1}, "2026-05": {"recommend": 4, "neutral": 3}, "2026-06": {"recommend": 11, "neutral": 9, "warn": 3, "compare": 3}, "2026-07": {"recommend": 7, "neutral": 9, "warn": 1}, "2026-08": {"recommend": 6, "neutral": 11}, "2026-09": {"recommend": 6, "neutral": 5, "compare": 2, "warn": 1}, "2026-10": {"recommend": 5, "warn": 2, "neutral": 3}}, "quotes": [{"id": 804291777, "item_id": 50268236, "url": "https://news.ycombinator.com/item?id=50268236", "by": "fultonn", "date": "2026-10-28", "stance": "recommend", "quote": "I use granite and qwen together with a manually curated set of books, articles, etc. to get work done while disconnected.", "excerpt": "Cool project. I have some property that doesn't really get cell reception in the summer, and even in the winter the reception is at-best spotty. I use granite and qwen together with a manually curated set of books, articles, etc. to get work done while disconnected. I tend to update the set of resources before leaving town because I'm generally intentionally disconnecting to focus on deep work. Building a workflow th", "context": "works offline at a remote property", "use_case": "offline work with local models", "switched_from": "", "story": "Survival LLM: AI assistant and emergency library that runs on your own computer"}, {"id": 803562657, "item_id": 50222666, "url": "https://news.ycombinator.com/item?id=50222666", "by": "chrisinnes", "date": "2026-10-24", "stance": "recommend", "quote": "Preferred Qwen.", "excerpt": "I used Kimi 2 (or 2.5) with my Openclaw and enjoyed it there so took out a subscription to Kimi 3.0 when it came out. Found it functional but boring. Preferred Qwen. Am not going to try the latest Kimi but it seems to be giving patchy results.", "context": "", "use_case": "LLM with OpenClaw", "switched_from": "Kimi", "story": "Kimi K3.1 Pro and K3.1 Pro Highspeed"}, {"id": 802560322, "item_id": 50160020, "url": "https://news.ycombinator.com/item?id=50160020", "by": "mlcruz", "date": "2026-10-20", "stance": "neutral", "quote": "even if you can easily run ds or qwen locally to get very similar results", "excerpt": "I think we keep getting used to intelligence improvements and the goal posts are always moving forward. I remember that close to an year ago i felt the same way about opus 4.5. And even if you can easily run ds or qwen locally to get very similar results to that, we moved to 5.5 being the new target. Deepseek can do 90% of the job but a claude reviwer/spec writer is a brutal improvement.", "context": "", "use_case": "running locally", "switched_from": "", "story": "MiniMax M3.1 Flash"}, {"id": 802356163, "item_id": 50147260, "url": "https://news.ycombinator.com/item?id=50147260", "by": "Perenti", "date": "2026-10-19", "stance": "warn", "quote": "all the models seem to love writing uncompileable code, and ending up in fearsome knots over it", "excerpt": "I love Free Pascal. I've tried using claude/codex/qwen to write in it, but all the models seem to love writing uncompileable code, and ending up in fearsome knots over it. It's my number one \"knock something together that just works\" language for humans though.", "context": "", "use_case": "writing Free Pascal code", "switched_from": "", "story": "Free Pascal compiler now supports dynamic packages"}, {"id": 801622689, "item_id": 50101418, "url": "https://news.ycombinator.com/item?id=50101418", "by": "thatsabadlook", "date": "2026-10-15", "stance": "recommend", "quote": "Even the dated qwen models from almost a year ago 1/10th the size are a better user experience", "excerpt": "Inkling last I used it was pretty terrible. I've never paid for a model that got stuck in a token puke loop and couldn't do tool calls until that day. Even the dated qwen models from almost a year ago 1/10th the size are a better user experience than the inkling model I piloted. Hoping they improve, sincerely, but not a good experience at all. What does it matter what country an open weight model comes from? Please d", "context": "", "use_case": "", "switched_from": "", "story": "LFM2.5-24B-A2B: Built for fast on-device agents"}, {"id": 801586578, "item_id": 50099161, "url": "https://news.ycombinator.com/item?id=50099161", "by": "dv35z", "date": "2026-10-15", "stance": "neutral", "quote": "still be able to chit-chat with Qwen model", "excerpt": "I used OpenCode to help me install ollama, Qwen local LLM, and connect OpenCode to it. The computer is very slow (old Dell Optiplex running Mint Debian Linux), but it worked! So cool to turn off the network, and still be able to chit-chat with Qwen model (\"Tell me a joke about...\").", "context": "", "use_case": "offline local chat", "switched_from": "", "story": "OpenCode 2.0"}, {"id": 800646912, "item_id": 50040432, "url": "https://news.ycombinator.com/item?id=50040432", "by": "kristopolous", "date": "2026-10-11", "stance": "neutral", "quote": "(well it's qwen with toolcalls)", "excerpt": "I've got something similar to this using cheap ONVIF security cameras and kasa smart bulbs. I have an nvidia card process the capture and for the smart speaker I use just a microphone and speaker hooked up to a normal computer. The architecture is spread across many machines and it's nowhere close to this realtime. Also the video capture is fairly lousy because the lighting and resolution and coverage isn't great. Th", "context": "", "use_case": "home camera and lighting automation", "switched_from": "", "story": "The Lightbulb Computer"}, {"id": 800638737, "item_id": 50039921, "url": "https://news.ycombinator.com/item?id=50039921", "by": "sachaa", "date": "2026-10-11", "stance": "warn", "quote": "the perf of LLM models like Qwen are not ideal for local execution", "excerpt": "This is a great way to start but the perf of LLM models like Qwen are not ideal for local execution. I have adapted Laya (pure decision model) to run in a browser and I am able to get responses under 200ms. Give it a try: https://wexare-ai.github.io/browser-laya/", "context": "", "use_case": "local LLM", "switched_from": "", "story": "Build your own decision model"}, {"id": 800419120, "item_id": 50026195, "url": "https://news.ycombinator.com/item?id=50026195", "by": "manmal", "date": "2026-10-09", "stance": "recommend", "quote": "a BF16 (unquantized) Qwen beats a Q8 of double its size for decisions", "excerpt": "My biggest learning after some experiments - a BF16 (unquantized) Qwen beats a Q8 of double its size for decisions. I guess that\u2019s the reason Kev switched to 4B BF16, from the original 8B version. Isn\u2019t it interesting that quantization seems to mess with decision accuracy?", "context": "", "use_case": "decision tasks", "switched_from": "", "story": "Microsoft-Decision-1, our model for fast decision-making"}, {"id": 799001152, "item_id": 49937572, "url": "https://news.ycombinator.com/item?id=49937572", "by": "verdverm", "date": "2026-10-02", "stance": "recommend", "quote": "It's my favorite model family to interact with", "excerpt": "I use all the qwen! It's my favorite model family to interact with, it's prose is the best imo, it makes me laugh from time-to-time (like when it said it would \"crib\" some code from another project, lul) I currently have qwen-flash working on an NES emulator harness so qwen-little can play my first RPG (ff1) (tho I have used all the others I mentioned, happenstance I'm using qwen this iteration/task)", "context": "", "use_case": "coding and agents", "switched_from": "", "story": "One month coding with GLM 5.3 Flash"}, {"id": 798591971, "item_id": 49911998, "url": "https://news.ycombinator.com/item?id=49911998", "by": "jeff_ciesielski", "date": "2026-09-30", "stance": "compare", "quote": "on my setup I found it a bit slow for my use case, and the benefits weren't strong enough to outweigh the slowdown", "excerpt": "Howdy folks. I was inspired by the really neat \"Jev in 25 lines of Python\" post a few days ago (and System One models in general), and after playing with the idea I tried whipping up something a little more complex, and Lichen is what popped out: https://github.com/Mushroom-Systems/lichen My question after reading that post (and I know I wasn't the only one; antirez came right out and said it in the comments) was whe", "context": "", "use_case": "local model", "switched_from": "", "story": "Lichen \u2013 A local, BOY model system1 (Jev) server with image support"}, {"id": 798349120, "item_id": 49896820, "url": "https://news.ycombinator.com/item?id=49896820", "by": "system2", "date": "2026-09-29", "stance": "recommend", "quote": "I also use QWEN and GLM for many of the api requiring projects", "excerpt": "Chinese model pressure. Many of my SWE friends switched to Chinese models. I also use QWEN and GLM for many of the api requiring projects and dropped OpenAI and Anthropic. The only reason was the cost. EDIT: I love getting downvoted by openai and anthropic employees or their bots.", "context": "", "use_case": "API-based projects", "switched_from": "", "story": "GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price"}, {"id": 797999985, "item_id": 49874999, "url": "https://news.ycombinator.com/item?id=49874999", "by": "babamogambo88", "date": "2026-09-28", "stance": "neutral", "quote": "I use this coding agent with qwen 2.5b", "excerpt": "https://github.com/STRAW-HAT-DEV/monkeyDcode.git I use this coding agent with qwen 2.5b and sometimes other models from open router.. as a one piece fan lol this looks nice with all the anime stuff .. but the work it does is also promising .. I don't know if it has any issues or something .. but I feel great using it .. it almost does all the tasks I ask it to do ..", "context": "", "use_case": "model for coding agent", "switched_from": "", "story": "Ask HN: Which model do you use for work?"}, {"id": 797901168, "item_id": 49868823, "url": "https://news.ycombinator.com/item?id=49868823", "by": "dsjakupov", "date": "2026-09-27", "stance": "neutral", "quote": "First, I used qwen without any ChatML-like syntax and it continued speaking and speaking", "excerpt": "Oh, I had similar case. First, I used qwen without any ChatML-like syntax and it continued speaking and speaking, then I used <|im_start|>/<|im_end|> to control it somehow", "context": "", "use_case": "local LLM", "switched_from": "", "story": "\"As a Language Model\": Chat Template Switches LLM Self-Referential Voice"}, {"id": 796414736, "item_id": 49775921, "url": "https://news.ycombinator.com/item?id=49775921", "by": "d2kx", "date": "2026-09-20", "stance": "recommend", "quote": "God I love the Qwen team. Easily the most diverse set of models from all the Chinese labs.", "excerpt": "God I love the Qwen team. Easily the most diverse set of models from all the Chinese labs. Only Gemini/DeepMind comes close.", "context": "", "use_case": "", "switched_from": "", "story": "Qwen Image 2.1"}, {"id": 796128082, "item_id": 49758005, "url": "https://news.ycombinator.com/item?id=49758005", "by": "HoldOnAMinute", "date": "2026-09-18", "stance": "warn", "quote": "it can do the work just fine - like creating a script - but it just outputs it, rather than placing it into a file.", "excerpt": "No - Here is what I mean. When I use Claude Code locally with Ollama/Quen, it can do the work just fine - like creating a script - but it just outputs it, rather than placing it into a file. So far I can't convince it to actually WRITE to a file. With Claude Code, it just works. Recursing around the file system, it just does it. I can't get that working locally.", "context": "", "use_case": "local coding model", "switched_from": "", "story": "Ask HN: What happens when Claude Code goes away?"}, {"id": 796093971, "item_id": 49755873, "url": "https://news.ycombinator.com/item?id=49755873", "by": "mmastrac", "date": "2026-09-18", "stance": "compare", "quote": "I ran the same evals against a Qwen36 and it clearly lost to both of them", "excerpt": "If you want to try a _legit_ Jev implementation that matches (at least in my evals), the vLLM patch to turn DiffusionGemma into Jev is available. On my DGX Spark I get very similar latency numbers, and it matches my evals + or - a few points on each test (DG wins some, Jev wins some, both show low confidence when wrong). I ran the same evals against a Qwen36 and it clearly lost to both of them, so you are leaving bot", "context": "", "use_case": "evals", "switched_from": "", "story": "OpenJev"}, {"id": 795986592, "item_id": 49749162, "url": "https://news.ycombinator.com/item?id=49749162", "by": "tolugenius", "date": "2026-09-18", "stance": "recommend", "quote": "one thing I love with Qwen is it comes a much larger range of sizes so I can experiment which extremely small llms.", "excerpt": "Curious if or when we'll see the Qwen4 series, one thing I love with Qwen is it comes a much larger range of sizes so I can experiment which extremely small llms.", "context": "", "use_case": "experimenting with small LLMs", "switched_from": "", "story": "Qwen 3.8 Omni Flash"}, {"id": 795348993, "item_id": 49709312, "url": "https://news.ycombinator.com/item?id=49709312", "by": "stevyhacker", "date": "2026-09-15", "stance": "neutral", "quote": "It runs locally by default with Qwen LLMs as the main engine.", "excerpt": "I\u2019m building LokalBot https://github.com/stevyhacker/lokalbot because I keep forgetting where I heard something or what we agreed on in a call. It\u2019s an open-source Mac app that transcribes meetings and can save screen context, so I can search through it later. It runs locally by default with Qwen LLMs as the main engine. I\u2019m paying even less attention in meetings now, though. \u201cI\u2019ll check the summary later\u201d has become", "context": "", "use_case": "local transcription summaries", "switched_from": "", "story": "Ask HN: What are you working on? (September 2026)"}, {"id": 795045376, "item_id": 49690336, "url": "https://news.ycombinator.com/item?id=49690336", "by": "noir_lord", "date": "2026-09-14", "stance": "recommend", "quote": "Qwen (at least for the things I use it for) edges just ahead", "excerpt": "I did, Muse came close but what impressed me about Qwen is it\u2019ll push back if it thinks it\u2019s right even when it isn\u2019t, I can work with that, Muse tended to flip between states too easily/too much. It is a good model but Qwen (at least for the things I use it for) edges just ahead, it seems much better at the \u201crip this apart, suggest improvements, touch nothing\u201d use case where I can use it as a second set of eyes, I d", "context": "runs local models on a 7900XTX", "use_case": "local code review", "switched_from": "", "story": "Astra and Fable still hack on simple variants of alignment evals from 2025"}, {"id": 794058496, "item_id": 49628656, "url": "https://news.ycombinator.com/item?id=49628656", "by": "tmzt", "date": "2026-09-09", "stance": "neutral", "quote": "using Qwen and Gemma models", "excerpt": "I've been experimenting with using J-Space (and final hidden layers) to extract the semantic meaning of words to improve TTS output accuracy using Qwen and Gemma models. My goal is to either map to an alternative token set where heteronyms are preserved or to output parenthesized IPA annotations for ambiguous words (with standard tokens). It's interesting to me that LLMs preserve this data throughout their processing", "context": "", "use_case": "improving TTS output accuracy", "switched_from": "", "story": "Extracting Steering Vectors from J space"}, {"id": 793835361, "item_id": 49614710, "url": "https://news.ycombinator.com/item?id=49614710", "by": "moffkalast", "date": "2026-09-08", "stance": "recommend", "quote": "I don't think any Mistral model can match an open 30B-sized Qwen from a year ago", "excerpt": "I don't think any Mistral model can match an open 30B-sized Qwen from a year ago, so right now they're not really a competitive alternative to anything at all. Except Voxtral perhaps, but that's very niche.", "context": "", "use_case": "", "switched_from": "", "story": "Mistral raises \u20ac3B"}, {"id": 793768576, "item_id": 49610536, "url": "https://news.ycombinator.com/item?id=49610536", "by": "lwarfield", "date": "2026-09-08", "stance": "neutral", "quote": "I self computed a j lens for the 27b version of the qwen model.", "excerpt": "If the author would like, I self computed a j lens for the 27b version of the qwen model. I used it for my own exploration in this area, and can share it if you want.", "context": "", "use_case": "interpretability exploration", "switched_from": "", "story": "Extracting Steering Vectors from J space"}, {"id": 792444849, "item_id": 49527803, "url": "https://news.ycombinator.com/item?id=49527803", "by": "aka-rider", "date": "2026-09-01", "stance": "recommend", "quote": "I use local Qwen to babysit these sessions, to make initial investigation, a repro case, and file a ticket.", "excerpt": "All very good questions. Agents are actively destroy QA gates in many ways, usually by cheating (\"the test is buggy, not my changes\" \u2014 changes the test), or just rot QA slowly by writing buggy overcomplicated tests What works for me 10/10 is fuzzing and my own constant usage. For this project specifically (text editor), I asked LLM to create human-like fuzzing session, it sends keystrokes like: \"the user is searching", "context": "", "use_case": "babysitting fuzzing sessions", "switched_from": "", "story": "I used Fable to rewrite 65kLoC of Go in Rust. It cost $400"}, {"id": 792223392, "item_id": 49513962, "url": "https://news.ycombinator.com/item?id=49513962", "by": "irishcoffee", "date": "2026-08-31", "stance": "recommend", "quote": "I use qwen at my house, it\u2019s pretty great.", "excerpt": "I think I\u2019m just going to find the best model I\u2019m authorized to use (of a few of them) and get on with it. You\u2019re much more passionate than I am about Chinese models, I don\u2019t really care about the topic much one way or the other. I show up to work to get paid and go home. I use qwen at my house, it\u2019s pretty great. Not gonna make problems for myself at work that I don\u2019t need to make, and I sure don\u2019t have any interest", "context": "", "use_case": "personal local use", "switched_from": "", "story": "vLLM v0.28.0"}, {"id": 792161104, "item_id": 49510069, "url": "https://news.ycombinator.com/item?id=49510069", "by": "inventor7777", "date": "2026-08-31", "stance": "neutral", "quote": "I use Qwen/GPT-OSS on my Mac Studio via llama.cpp server.", "excerpt": "One use case I find cost effective is using it as a voice assistant for Home Assistant. API pricing on models is very weird compared to the normal chats, so I use Qwen/GPT-OSS on my Mac Studio via llama.cpp server.", "context": "", "use_case": "Home Assistant voice assistant", "switched_from": "", "story": "Apple caught off guard by AI demand for Mac Mini and Mac Studio"}, {"id": 791988562, "item_id": 49499285, "url": "https://news.ycombinator.com/item?id=49499285", "by": "conikeec", "date": "2026-08-30", "stance": "neutral", "quote": "send simpler work from Codex (sol xhigh) to cheaper models (Luna, Qwen)", "excerpt": "Longer explanation: https://modiqo.github.io/spewer/ For the past few months, we have been using an internal tool to send simpler work from Codex (sol xhigh) to cheaper models (Luna, Qwen). We called it delegate, which described the function but never felt like a real name. Tidy'd it up and released it as Spewer. I hope any you find it useful as we did.", "context": "", "use_case": "cheaper model for simple tasks", "switched_from": "", "story": "Show HN: Spewer \u2013 Delegate Codex/Claude tasks to cheaper models"}, {"id": 791115040, "item_id": 49444690, "url": "https://news.ycombinator.com/item?id=49444690", "by": "SturgeonsLaw", "date": "2026-08-26", "stance": "recommend", "quote": "I use Qwen/Claude for coding and Gemini/Gemma to draft text intended for a human.", "excerpt": "I use Qwen/Claude for coding and Gemini/Gemma to draft text intended for a human. To me that feels like each model playing to their strengths.", "context": "", "use_case": "coding", "switched_from": "", "story": "Analyzing student votes across AI models for college essay help"}, {"id": 790890785, "item_id": 49430674, "url": "https://news.ycombinator.com/item?id=49430674", "by": "ciefa", "date": "2026-08-25", "stance": "recommend", "quote": "the rest Qwen/Kimi/Deepseek can do", "excerpt": "I switched to GPT 5.6 Sol for the hardest tasks, the rest Qwen/Kimi/Deepseek can do. So far not looking back to Anthropic. They lost me.", "context": "", "use_case": "routine tasks", "switched_from": "", "story": "Anthropic's best AI model struggles to attract users as cheaper tools thrive"}, {"id": 790631876, "item_id": 49414492, "url": "https://news.ycombinator.com/item?id=49414492", "by": "llm_nerd", "date": "2026-08-24", "stance": "neutral", "quote": "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", "excerpt": ">You don't feel they want to give you a dependable service for an, albeit premium, price. It's a constant bargaining game We all know that the subscription prices are not at all sustainable for these providers. You all do, right? Yes, they're struggling to segment the market and find a way to make money, and that basically relies upon emptying the pockets of whales. As someone enjoying a hilariously subsidized Max pl", "context": "", "use_case": "", "switched_from": "", "story": "Anthropic's best AI model struggles to attract users as cheaper tools thrive"}, {"id": 790526656, "item_id": 49407916, "url": "https://news.ycombinator.com/item?id=49407916", "by": "synthrakx", "date": "2026-08-23", "stance": "neutral", "quote": "I have been using open source LLMs locally for the past 5 months like Qwen, Llama, DeepSeek and others", "excerpt": "I have been using open source LLMs locally for the past 5 months like Qwen, Llama, DeepSeek and others, and I have also noticed that current local models feel significantly dumber than closed source commercial models like ChatGPT, Claude, and Gemini. One main reason I think is the amount, variety, and quality of original authentic data on which they are being trained on, and also the training method plays a significa", "context": "", "use_case": "local LLM use", "switched_from": "", "story": "Why your local LLM feels dumber than it is"}, {"id": 790266402, "item_id": 49391650, "url": "https://news.ycombinator.com/item?id=49391650", "by": "100percentjake", "date": "2026-08-21", "stance": "neutral", "quote": "Qwen running locally on a 64GB M1 Max Macbook Pro to help save my ever-quickly shrinking Codex credits", "excerpt": "Neat stuff. I use Hermes with the $20/mo ChatGPT Codex integration, myself, as well as Qwen running locally on a 64GB M1 Max Macbook Pro to help save my ever-quickly shrinking Codex credits. It's been wonderful for stuff like, say, giving Hermes an account on my Home Assistant server (after taking a backup) and having it make a bunch of configurations, or draw conclusions based on historical sensor data (trying to de", "context": "", "use_case": "local model", "switched_from": "", "story": "A self hosted AI software factory"}, {"id": 790258496, "item_id": 49391156, "url": "https://news.ycombinator.com/item?id=49391156", "by": "nater5000", "date": "2026-08-21", "stance": "neutral", "quote": "The Qwen models are decent, but they don't come close to the full Claude experience I've come to expect.", "excerpt": "I have a similar setup to the OP (although I'm not sure if I'd call it a \"software factory\"), and I utilize a local model for some aspects of the setup. Specifically, I have a single RTX 3090 Ti with 24GB VRAM on my (main) home server, and I've been primarily running Qwen 3.6 35B A3B out of it via Ollama (I just switched to Qwen 3.8 27B, though, and I've also tried other Qwen models as well as Gemma models). The Qwen", "context": "runs a home server with an RTX 3090 Ti", "use_case": "local model for specific tasks", "switched_from": "", "story": "A self hosted AI software factory"}, {"id": 789790769, "item_id": 49361923, "url": "https://news.ycombinator.com/item?id=49361923", "by": "nozzlegear", "date": "2026-08-19", "stance": "neutral", "quote": "it's still a problem with the Qwen and Deepseek models I use now", "excerpt": "Could be a symptom of which languages we use and which languages are in their training sets. I primarily write F#, which is a functional language that's compatible with the object-oriented C#. When AI tries to write code in my F# codebases, it almost always ignores the existing patterns/styles and tries to write it like an OO language. It has a weird desire to use mutability instead of the default immutability, for l", "context": "F# developer", "use_case": "writing F# code", "switched_from": "Claude", "story": "AI usage patterns in software teams"}, {"id": 788901152, "item_id": 49306322, "url": "https://news.ycombinator.com/item?id=49306322", "by": "arjie", "date": "2026-08-14", "stance": "neutral", "quote": "I use Qwen mostly for vision in my personal assistant.", "excerpt": "Most of us aren\u2019t deploying these in general purpose use-cases. E.g. I use Qwen mostly for vision in my personal assistant. I have an eval set for that. Pretty much each of my use cases has a pre-computed problem set. Likewise I have a tool-use eval set and a browser-use eval set. I use frontier models for most interactive tasks that are not home assistant or background agents. I can imagine someone could build evals", "context": "", "use_case": "vision in personal assistant", "switched_from": "", "story": "Qwen3.8-27B"}, {"id": 788813040, "item_id": 49300815, "url": "https://news.ycombinator.com/item?id=49300815", "by": "arjie", "date": "2026-08-14", "stance": "neutral", "quote": "I use the Qwens as a vision model for my DeepSeek V4 Flashes to handle.", "excerpt": "I use the Qwens as a vision model for my DeepSeek V4 Flashes to handle. But the Qwens run on old RTX A6000 Ampere. Does anyone know if there's any news about INT4/AWQ quants for the RTX A6000?", "context": "runs local GPUs", "use_case": "vision model", "switched_from": "", "story": "Qwen 3.8 27B is out: open weights, best local dense model yet"}, {"id": 788810610, "item_id": 49300663, "url": "https://news.ycombinator.com/item?id=49300663", "by": "simplyluke", "date": "2026-08-14", "stance": "recommend", "quote": "actually prefer the open weight models in most day to day usage", "excerpt": "I'm convinced a lot of the anti-open-weight model comments at this point are inorganic traffic - there's trillions in investor money riding on a world where these models aren't cheap commodities. Having actually used things like the recent GLM, Kimi, and Qwen I think any edge the labs have is marginal at most and actually prefer the open weight models in most day to day usage. Anthropic's recent releases are wordy to", "context": "", "use_case": "day to day usage", "switched_from": "", "story": "Qwen3.8-27B"}, {"id": 788800640, "item_id": 49300040, "url": "https://news.ycombinator.com/item?id=49300040", "by": "NitpickLawyer", "date": "2026-08-14", "stance": "recommend", "quote": "Qwen models are impressively good for what they are, are \"good enough\" for plenty tasks, can be ran locally on decently priced hardware", "excerpt": "> Beats Opus 4.7 Max I'm a huge open model fan, and have used them since forever, even have daily drivers for on-prem dev, but no. They do not beat opus on real-world usage. Qwen models are impressively good for what they are, are \"good enough\" for plenty tasks, can be ran locally on decently priced hardware, and so on. They certainly have their uses, and the field in general has advanced faster than my early expecta", "context": "open model fan with on-prem dev", "use_case": "local on-prem dev", "switched_from": "", "story": "Qwen3.8-27B"}, {"id": 788711425, "item_id": 49294464, "url": "https://news.ycombinator.com/item?id=49294464", "by": "Wpnx330", "date": "2026-08-13", "stance": "neutral", "quote": "put a few qwen models for vision and embedding purposes on my original DGX Spark", "excerpt": "$0. but I use at least $8,000 worth of monthly tokens. I used to have: $200 Claude-max $10 github copilot $20 chatgpt $20 Cursor --- then, I spent roughly $5000 in tokens for a project which turned out to be a dud. Figured, inference wasn't going anywhere as its now 80% of my day. So, I built a 4x DGX Spark cluster and put GLM5.2 on it. Then put a few qwen models for vision and embedding purposes on my original DGX S", "context": "", "use_case": "vision and embeddings", "switched_from": "", "story": "Ask HN: How much money do you spend monthly on subscriptions for AI models?"}, {"id": 788028181, "item_id": 49251761, "url": "https://news.ycombinator.com/item?id=49251761", "by": "danielnexaaly", "date": "2026-08-10", "stance": "neutral", "quote": "Codex CLI, Claude Code, Qwen", "excerpt": "Location: Herning, Denmark (UTC+2) Remote: Yes Willing to relocate: Not listing relocation availability at this time Technologies: Linux, JavaScript/Node.js, Git, Docker, HTTP/APIs, browser automation, MCP, Caddy, Stripe, Codex CLI, Claude Code, Qwen Resume/CV: https://www.linkedin.com/in/daniel-abildgaard-376032190/ Portfolio: https://nexaaly.com/bug-hunt/sample/ and https://github.com/AccessNexaaly Email: access@ne", "context": "AI-assisted technical operator", "use_case": "", "switched_from": "", "story": "Ask HN: Who wants to be hired? (August 2026)"}, {"id": 780461297, "item_id": 48778831, "url": "https://news.ycombinator.com/item?id=48778831", "by": "verdverm", "date": "2026-07-03", "stance": "recommend", "quote": "It can be considered SOTA within is size category. Very useful for many things.", "excerpt": "It can be considered SOTA within is size category. Very useful for many things. You still want access to big models, I recommend OpenCode Go if you want to stay with open models.", "context": "", "use_case": "local LLM", "switched_from": "", "story": "Jamesob's guide to running SOTA LLMs locally"}, {"id": 780156032, "item_id": 48759752, "url": "https://news.ycombinator.com/item?id=48759752", "by": "jadbox", "date": "2026-07-02", "stance": "recommend", "quote": "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.", "excerpt": "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.", "context": "", "use_case": "local daily driver model", "switched_from": "", "story": "Kimi K2.7 Code is generally available in GitHub Copilot"}, {"id": 779566403, "item_id": 48722900, "url": "https://news.ycombinator.com/item?id=48722900", "by": "blopker", "date": "2026-06-29", "stance": "warn", "quote": "Qwen in particular seems to work at first, then often gets stuck in thought loops when used for actual work.", "excerpt": "I've been working with local models for the past year. There's so many possibilities, but I don't think coding is one. Coding requires so many layers beyond inference; I spent so much time trying to replicate what Claude Code does end to end locally. Understanding all the layers and keeping up with the advancements feels like a slog. Even this article messes up and misunderstands what some of the settings are doing. ", "context": "", "use_case": "local coding", "switched_from": "", "story": "Qwen 3.6 27B is the sweet spot for local development"}, {"id": 778266354, "item_id": 48641647, "url": "https://news.ycombinator.com/item?id=48641647", "by": "yenihesap", "date": "2026-06-23", "stance": "recommend", "quote": "I recommend DeepSeek, GLM/Z-AI or Qwen.", "excerpt": "Just use one of the alternatives, there are so many. I recommend DeepSeek, GLM/Z-AI or Qwen. They are cheaper and offer competitive performance to Anthropic.", "context": "", "use_case": "coding assistant alternative", "switched_from": "", "story": "Ask HN: Anthropic banned me from using Claude Code and I don't know what to do"}, {"id": 777377792, "item_id": 48586112, "url": "https://news.ycombinator.com/item?id=48586112", "by": "Imustaskforhelp", "date": "2026-06-18", "stance": "recommend", "quote": "Qwen models are generally the best if you want local models in the 100b max size.", "excerpt": "Qwen models are generally the best if you want local models in the 100b max size. I am not sure but I recommend looking at the size of deepseek flash as well to see if it can work for you, perhaps GLM 5 Air model is also around this size. and if you want something at the larger models scale, then you have glm 5.2, kimi k2.7,mimo v2.5 pro and deepseek v4", "context": "", "use_case": "local models", "switched_from": "", "story": "Ask HN: Whats the best and small open source model?"}, {"id": 777018560, "item_id": 48563660, "url": "https://news.ycombinator.com/item?id=48563660", "by": "qudat", "date": "2026-06-16", "stance": "recommend", "quote": "A 24gb vram, 5 year old titan run 27b at 30t/s and the results are good.", "excerpt": "I feel like the claims come from wildly different personas and use cases. A 24gb vram, 5 year old titan run 27b at 30t/s and the results are good. I use sonnet and opus at my day job and they are more capable but I can still get the same out of qwen, I just need to be mindful of ctx", "context": "uses Sonnet and Opus at day job", "use_case": "local inference", "switched_from": "", "story": "Running local models is good now"}, {"id": 776725328, "item_id": 48545333, "url": "https://news.ycombinator.com/item?id=48545333", "by": "major505", "date": "2026-06-15", "stance": "recommend", "quote": "I use Owen on my MacBook m1 (16gb) daily, running inside Ollama. Works well.", "excerpt": "Yes. I use Owen on my MacBook m1 (16gb) daily, running inside Ollama. Works well. Is not particularly fast, and I need to create a custom imagem that sets the temperature of the model to zero starting, so I don't get over creative with its bullshit, but it works reasonable week.", "context": "", "use_case": "local daily coding", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}]}