{"slug": "gpt-oss", "name": "GPT-OSS", "category": "ai", "category_name": "AI models and assistants", "description": "OpenAI's open-weight large language models, released in 20B and 120B parameter sizes.", "recommenders": 15, "commenters": 31, "warners": 2, "score": 13.102, "rising": -0.93, "first": "2025-11-01", "latest": "2026-10-30", "why": "HN commenters like it for running locally on laptops and home servers, for fast output, and for agentic and coding work.", "why_is_quote": false, "has_verdict": true, "sentiment": {"recommend": 16, "neutral": 13, "compare": 3, "warn": 2}, "spark": [3, 2, 2, 2, 0, 1, 1, 2, 0, 1, 0, 2], "listed": true, "loved": ["Runs locally", "Speed"], "caveat": "20B mistakes", "short_quote": "Once I did that it started functioning extremely well, and it's the main model I use for my homemade agents.", "icon": true, "domain": null, "github_repo": null, "official_url": "https://openai.com/open-models/", "pricing": "free", "aliases": ["gpt-oss-20b", "gpt-oss-120b", "GPT-OSS", "gpt-oss", "GPT-OSS 20B", "gpt-oss 120b", "GPT-OSS 120B", "GPT-OSS 120b", "GPT-oss-120B", "gpt-oss-20B", "GPT OSS 120B", "GPT-OSS-120B", "GPT OSS", "GPT OSS 20B"], "verdict": [{"text": "HN commenters like it for running locally on laptops and home servers,", "cites": [{"id": 804880544, "item_id": 50305034, "url": "https://news.ycombinator.com/item?id=50305034", "by": "Breza", "date": "2026-10-30", "stance": "recommend", "quote": "I find myself firing up GPT-0SS-20b on my laptop for smaller tasks", "excerpt": "I agree with you. I use LLMs for a lot of different things. But increasingly, I find myself firing up GPT-0SS-20b on my laptop for smaller tasks. And I know there are better models I just haven't tried yet.", "context": "", "use_case": "local LLM for small tasks", "switched_from": "", "story": "Anthropic S-1"}, {"id": 755262544, "item_id": 47203909, "url": "https://news.ycombinator.com/item?id=47203909", "by": "carlio", "date": "2026-02-28", "stance": "recommend", "quote": "it can run a 120b model (gpt-oss:120b) just fine.", "excerpt": "An AMD AI max+ 395 - I use the one from frame.work (https://frame.work/de/en/desktop) with 128GB unified RAM and it can run a 120b model (gpt-oss:120b) just fine. See Wendel's review here - https://www.youtube.com/watch?v=L-xgMQ-7lW0 There are other mini-pc manufacturers, the mainboard is the important part.", "context": "", "use_case": "local LLM inference", "switched_from": "", "story": "Qwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers"}, {"id": 732478098, "item_id": 45779881, "url": "https://news.ycombinator.com/item?id=45779881", "by": "Greenpants", "date": "2025-11-01", "stance": "recommend", "quote": "I rarely ever use ChatGPT anymore. Love that all data stays at home.", "excerpt": "I got a personal Mac Studio M4 Max with 128GB RAM for a silent, relatively power-efficient yet powerful home server. It runs Ollama + Open WebUI with GPT-OSS 120b as well as GLM4.5-Air (default quantisations). I rarely ever use ChatGPT anymore. Love that all data stays at home. I connect remotely only via VPN (my phone enables this automatically via Tasker). I'm 50% brainstorming ideas with it, asking critical questi", "context": "", "use_case": "local brainstorming and coding", "switched_from": "ChatGPT", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}, {"id": 733134305, "item_id": 45820894, "url": "https://news.ycombinator.com/item?id=45820894", "by": "EagnaIonat", "date": "2025-11-05", "stance": "recommend", "quote": "gpt-oss. 20b or 120b depending on complexity of use cases.", "excerpt": "The more recent LLMs work fine on an M1 mac. Can't speak for Windows/Linux. There was even a recent release of Granite4 that runs on a Raspberry Pi. https://github.com/Jewelzufo/granitepi-4-nano For my local work I use Ollama. (M4 Max 128GB) - gpt-oss. 20b or 120b depending on complexity of use cases. - granite4 for speed and lower complexity (around the same as gpt20b).", "context": "M4 Max 128GB Mac user", "use_case": "local LLM work", "switched_from": "", "story": "Developers are choosing older AI models"}]}, {"text": " for fast output,", "cites": [{"id": 775981172, "item_id": 48498823, "url": "https://news.ycombinator.com/item?id=48498823", "by": "inventor7777", "date": "2026-06-11", "stance": "recommend", "quote": "runs GPT-OSS 20B at over 100 tok/sec", "excerpt": "5-10 sessions?! Wow, not sure how you keep track of those. For all my daily use + Codex Desktop and some local AIs in Ollama/LM Studio (which power my Home Assistant Voice Assistants) I have an Mac Studio M4 Max with 48GB of RAM (Which I bought early last year IIRC) It's a complete beast and runs GPT-OSS 20B at over 100 tok/sec. The GUI never slows down, and I run Sequoia still, so it is always smooth and polished. I", "context": "", "use_case": "local inference", "switched_from": "", "story": "Ask HN: What computer are you using for AI coding tools?"}, {"id": 761924928, "item_id": 47620308, "url": "https://news.ycombinator.com/item?id=47620308", "by": "evilduck", "date": "2026-04-02", "stance": "recommend", "quote": "gpt-oss-120b is over 600 tokens/s PP for all but one backend.", "excerpt": "In terms of ability, maybe, in terms of speed, it's not even close. Check out the Prompt Processing speeds between them: https://kyuz0.github.io/amd-strix-halo-toolboxes/ gpt-oss-120b is over 600 tokens/s PP for all but one backend. nemotron-3-super is at best 260 tokens/s PP. Comparing token generation, it's again like 50 tokens/sec vs 15 tokens/sec That really bogs down agentic tooling. Something needs to be catego", "context": "", "use_case": "local agentic inference", "switched_from": "", "story": "Google releases Gemma 4 open models"}, {"id": 742139857, "item_id": 46383741, "url": "https://news.ycombinator.com/item?id=46383741", "by": "sgk284", "date": "2025-12-25", "stance": "recommend", "quote": "We use Groq with gpt-oss-20b, which gives great results and only adds ~250ms to the processing pipeline.", "excerpt": "Reranking is definitely the way to go. We personally found common reranker models to be a little too opaque (can't explain to the user why this result was picked) and not quite steerable enough, so we just use another LLM for reranking. We open-sourced our impl just this week: https://github.com/with-logic/intent We use Groq with gpt-oss-20b, which gives great results and only adds ~250ms to the processing pipeline. ", "context": "", "use_case": "reranking search results", "switched_from": "", "story": "Using Vectorize to build an unreasonably good search engine in 160 lines of code"}]}, {"text": " and for agentic and coding work.", "cites": [{"id": 775033664, "item_id": 48439604, "url": "https://news.ycombinator.com/item?id=48439604", "by": "computerex", "date": "2026-06-07", "stance": "recommend", "quote": "It\u2019s highly capable in agentic settings, people use it for coding all the time.", "excerpt": "You are actually referring to open weight models, not open source. Gpt-OSS is an example of an open weight model. It\u2019s highly capable in agentic settings, people use it for coding all the time. My greater point remains. Models like the qwen variants, minimax, k2.5, glm models are available by American providers like AWS at a much cheaper price than api offerings from the big three LLM providers. Your point about Chin", "context": "", "use_case": "agentic coding", "switched_from": "", "story": "The OnlyFans Economy of American AI"}, {"id": 741833552, "item_id": 46364597, "url": "https://news.ycombinator.com/item?id=46364597", "by": "nekitamo", "date": "2025-12-23", "stance": "recommend", "quote": "it started functioning extremely well, and it's the main model I use for my homemade agents.", "excerpt": "GPT-oss-120B was also completely failing for me, until someone on reddit pointed out that you need to pass back in the reasoning tokens when generating a response. One way to do this is described here: https://openrouter.ai/docs/guides/best-practices/reasoning-tokens Once I did that it started functioning extremely well, and it's the main model I use for my homemade agents. Many LLM libraries/services/frontends don't", "context": "builds own agents", "use_case": "homemade agents", "switched_from": "", "story": "GLM-4.7: Advancing the Coding Capability"}, {"id": 770750021, "item_id": 48171876, "url": "https://news.ycombinator.com/item?id=48171876", "by": "zambelli", "date": "2026-05-17", "stance": "neutral", "quote": "GPT-OSS-120B for self-hosted coding or more reasoning-intensive agentic flows.", "excerpt": "I use Opus 4.7 for personal stuff (basically for everything), but have been considering gpt-5.5 given all I hear about it. At work I use 4.6 because we don't have 4.7 yet...zzz... I also do a LOT of personal/portfolio work with self-hosted models. Ministral-3-14B-Reasoning for validating concepts, MVPs, etc and some prod systems (punches above its weight class). Qwen3.6-35B-A3B for self-hosted coding (custom harness)", "context": "", "use_case": "self-hosted coding, agentic flows", "switched_from": "", "story": "Ask HN: What LLM models are you using and why?"}]}, {"text": " The most common complaint is that the 20B model makes frequent mistakes on harder tasks,", "cites": [{"id": 769439041, "item_id": 48089940, "url": "https://news.ycombinator.com/item?id=48089940", "by": "quacker", "date": "2026-05-10", "stance": "neutral", "quote": "GPT OSS 20B was usable but slow, and actually frequently made mistakes", "excerpt": "I could have used this article before I spent the weekend arriving to the same conclusion! Same laptop, and my contrived test was having it fix 50 or so lint errors in a small vibe-coded C++ repo. I wanted it to be able to handle a bunch of small tasks without getting stuck too often. GPT OSS 20B was usable but slow, and actually frequently made mistakes like adding or duplicating statements unnecessarily, listing th", "context": "", "use_case": "local lint fixing", "switched_from": "", "story": "Running local models on an M4 with 24GB memory"}, {"id": 732500112, "item_id": 45781257, "url": "https://news.ycombinator.com/item?id=45781257", "by": "embedding-shape", "date": "2025-11-01", "stance": "recommend", "quote": "for even moderately difficult programming tasks, 20B fails repeatedly while 120B gets it right on the first try", "excerpt": "> had better results with the 20b model, over the 120b model The difference of quality and accuracy of the responses between the two is vastly different though, if tok/s isn't your biggest priority, especially when using reasoning_effort \"high\". 20B works great for small-ish text summarization and title generation, but for even moderately difficult programming tasks, 20B fails repeatedly while 120B gets it right on t", "context": "", "use_case": "programming tasks", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}, {"id": 734912880, "item_id": 45932055, "url": "https://news.ycombinator.com/item?id=45932055", "by": "ticulatedspline", "date": "2025-11-14", "stance": "neutral", "quote": "I tried gpt-oss-20b (my go-to local) and it looks ok though not very accurate.", "excerpt": "This is cool, interesting to see how consistent some models are (both in success and failure) I tried gpt-oss-20b (my go-to local) and it looks ok though not very accurate. It decided to omit numbers. It also took 4500 tokens while thinking. I'd be interested in seeing it with some more token leeway as well as comparing two or more similar prompts. like using \"current time\" instead of \"${time}\" and being more prescri", "context": "", "use_case": "local LLM", "switched_from": "", "story": "AI World Clocks"}]}, {"text": " and some say the 120B model only worked well after setup fixes like passing back reasoning tokens.", "cites": [{"id": 750104032, "item_id": 46881502, "url": "https://news.ycombinator.com/item?id=46881502", "by": "nekitamo", "date": "2026-02-03", "stance": "recommend", "quote": "Once I did that it started functioning extremely well, and it's the main model I use for my homemade agents.", "excerpt": "This is a common problem for people trying to run the GPT-oss models themselves. Reposting my comment here: GPT-oss-120B was also completely failing for me, until someone on reddit pointed out that you need to pass back in the reasoning tokens when generating a response. One way to do this is described here: https://openrouter.ai/docs/guides/best-practices/reasoning-tokens Once I did that it started functioning extre", "context": "builds homemade agents", "use_case": "homemade agents", "switched_from": "", "story": "Qwen3-Coder-Next"}, {"id": 741833552, "item_id": 46364597, "url": "https://news.ycombinator.com/item?id=46364597", "by": "nekitamo", "date": "2025-12-23", "stance": "recommend", "quote": "it started functioning extremely well, and it's the main model I use for my homemade agents.", "excerpt": "GPT-oss-120B was also completely failing for me, until someone on reddit pointed out that you need to pass back in the reasoning tokens when generating a response. One way to do this is described here: https://openrouter.ai/docs/guides/best-practices/reasoning-tokens Once I did that it started functioning extremely well, and it's the main model I use for my homemade agents. Many LLM libraries/services/frontends don't", "context": "builds own agents", "use_case": "homemade agents", "switched_from": "", "story": "GLM-4.7: Advancing the Coding Capability"}, {"id": 733032305, "item_id": 45814519, "url": "https://news.ycombinator.com/item?id=45814519", "by": "embedding-shape", "date": "2025-11-04", "stance": "recommend", "quote": "The 20B is definitely worse than 120B for me in every case and scenario", "excerpt": "> But the 120b model has just as bad if not worse formatting issues, compared to the 20b one What runtime/tools are you using? Haven't been my experience at all, but I've also mostly used it via llama.cpp and my own \"coding agent\". It was slightly tricky to get the Harmony parsing in place and working correct, but once that's in place, I haven't seen any formatting issues at all? The 20B is definitely worse than 120B", "context": "", "use_case": "local coding assistant", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}]}], "praise": [{"label": "Runs locally", "summary": "Commenters run it on laptops, Macs and mini PCs, with the 120B model running fine on 128GB machines.", "quotes": [{"id": 804880544, "item_id": 50305034, "url": "https://news.ycombinator.com/item?id=50305034", "by": "Breza", "date": "2026-10-30", "stance": "recommend", "quote": "I find myself firing up GPT-0SS-20b on my laptop for smaller tasks", "excerpt": "I agree with you. I use LLMs for a lot of different things. But increasingly, I find myself firing up GPT-0SS-20b on my laptop for smaller tasks. And I know there are better models I just haven't tried yet.", "context": "", "use_case": "local LLM for small tasks", "switched_from": "", "story": "Anthropic S-1"}, {"id": 755262544, "item_id": 47203909, "url": "https://news.ycombinator.com/item?id=47203909", "by": "carlio", "date": "2026-02-28", "stance": "recommend", "quote": "it can run a 120b model (gpt-oss:120b) just fine.", "excerpt": "An AMD AI max+ 395 - I use the one from frame.work (https://frame.work/de/en/desktop) with 128GB unified RAM and it can run a 120b model (gpt-oss:120b) just fine. See Wendel's review here - https://www.youtube.com/watch?v=L-xgMQ-7lW0 There are other mini-pc manufacturers, the mainboard is the important part.", "context": "", "use_case": "local LLM inference", "switched_from": "", "story": "Qwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers"}, {"id": 790939041, "item_id": 49433690, "url": "https://news.ycombinator.com/item?id=49433690", "by": "super_mario", "date": "2026-08-25", "stance": "neutral", "quote": "I can run local LLMs like Gemma 3/4, gpt-oss 120b all day long", "excerpt": "I was in the same situation, I used maxed out 15'' M3 Max MacBook Pro docked to Studio Display closed on vertical stand behind the screen. It was fine for office work, but running local LLMs would definitely overheat it. The battery started degrading purely due to heat issues. And it was audible as well. I decided to get Mac Studio M4 Max, also all maxed out config and the cooling is so much better that I can run loc", "context": "", "use_case": "local LLM inference", "switched_from": "", "story": "New Mac Studio with M5 Max and M5 Ultra"}]}, {"label": "Speed", "summary": "Users report fast token generation locally and low latency on hosted providers like Groq.", "quotes": [{"id": 775981172, "item_id": 48498823, "url": "https://news.ycombinator.com/item?id=48498823", "by": "inventor7777", "date": "2026-06-11", "stance": "recommend", "quote": "runs GPT-OSS 20B at over 100 tok/sec", "excerpt": "5-10 sessions?! Wow, not sure how you keep track of those. For all my daily use + Codex Desktop and some local AIs in Ollama/LM Studio (which power my Home Assistant Voice Assistants) I have an Mac Studio M4 Max with 48GB of RAM (Which I bought early last year IIRC) It's a complete beast and runs GPT-OSS 20B at over 100 tok/sec. The GUI never slows down, and I run Sequoia still, so it is always smooth and polished. I", "context": "", "use_case": "local inference", "switched_from": "", "story": "Ask HN: What computer are you using for AI coding tools?"}, {"id": 761924928, "item_id": 47620308, "url": "https://news.ycombinator.com/item?id=47620308", "by": "evilduck", "date": "2026-04-02", "stance": "recommend", "quote": "gpt-oss-120b is over 600 tokens/s PP for all but one backend.", "excerpt": "In terms of ability, maybe, in terms of speed, it's not even close. Check out the Prompt Processing speeds between them: https://kyuz0.github.io/amd-strix-halo-toolboxes/ gpt-oss-120b is over 600 tokens/s PP for all but one backend. nemotron-3-super is at best 260 tokens/s PP. Comparing token generation, it's again like 50 tokens/sec vs 15 tokens/sec That really bogs down agentic tooling. Something needs to be catego", "context": "", "use_case": "local agentic inference", "switched_from": "", "story": "Google releases Gemma 4 open models"}, {"id": 742139857, "item_id": 46383741, "url": "https://news.ycombinator.com/item?id=46383741", "by": "sgk284", "date": "2025-12-25", "stance": "recommend", "quote": "We use Groq with gpt-oss-20b, which gives great results and only adds ~250ms to the processing pipeline.", "excerpt": "Reranking is definitely the way to go. We personally found common reranker models to be a little too opaque (can't explain to the user why this result was picked) and not quite steerable enough, so we just use another LLM for reranking. We open-sourced our impl just this week: https://github.com/with-logic/intent We use Groq with gpt-oss-20b, which gives great results and only adds ~250ms to the processing pipeline. ", "context": "", "use_case": "reranking search results", "switched_from": "", "story": "Using Vectorize to build an unreasonably good search engine in 160 lines of code"}]}, {"label": "Agents and coding", "summary": "Used as the main model for homemade agents and for self-hosted coding and tool use.", "quotes": [{"id": 741833552, "item_id": 46364597, "url": "https://news.ycombinator.com/item?id=46364597", "by": "nekitamo", "date": "2025-12-23", "stance": "recommend", "quote": "it started functioning extremely well, and it's the main model I use for my homemade agents.", "excerpt": "GPT-oss-120B was also completely failing for me, until someone on reddit pointed out that you need to pass back in the reasoning tokens when generating a response. One way to do this is described here: https://openrouter.ai/docs/guides/best-practices/reasoning-tokens Once I did that it started functioning extremely well, and it's the main model I use for my homemade agents. Many LLM libraries/services/frontends don't", "context": "builds own agents", "use_case": "homemade agents", "switched_from": "", "story": "GLM-4.7: Advancing the Coding Capability"}, {"id": 775033664, "item_id": 48439604, "url": "https://news.ycombinator.com/item?id=48439604", "by": "computerex", "date": "2026-06-07", "stance": "recommend", "quote": "It\u2019s highly capable in agentic settings, people use it for coding all the time.", "excerpt": "You are actually referring to open weight models, not open source. Gpt-OSS is an example of an open weight model. It\u2019s highly capable in agentic settings, people use it for coding all the time. My greater point remains. Models like the qwen variants, minimax, k2.5, glm models are available by American providers like AWS at a much cheaper price than api offerings from the big three LLM providers. Your point about Chin", "context": "", "use_case": "agentic coding", "switched_from": "", "story": "The OnlyFans Economy of American AI"}, {"id": 770750021, "item_id": 48171876, "url": "https://news.ycombinator.com/item?id=48171876", "by": "zambelli", "date": "2026-05-17", "stance": "neutral", "quote": "GPT-OSS-120B for self-hosted coding or more reasoning-intensive agentic flows.", "excerpt": "I use Opus 4.7 for personal stuff (basically for everything), but have been considering gpt-5.5 given all I hear about it. At work I use 4.6 because we don't have 4.7 yet...zzz... I also do a LOT of personal/portfolio work with self-hosted models. Ministral-3-14B-Reasoning for validating concepts, MVPs, etc and some prod systems (punches above its weight class). Qwen3.6-35B-A3B for self-hosted coding (custom harness)", "context": "", "use_case": "self-hosted coding, agentic flows", "switched_from": "", "story": "Ask HN: What LLM models are you using and why?"}]}, {"label": "120B quality", "summary": "The 120B model gets moderately difficult programming tasks right where the 20B fails.", "quotes": [{"id": 732500112, "item_id": 45781257, "url": "https://news.ycombinator.com/item?id=45781257", "by": "embedding-shape", "date": "2025-11-01", "stance": "recommend", "quote": "for even moderately difficult programming tasks, 20B fails repeatedly while 120B gets it right on the first try", "excerpt": "> had better results with the 20b model, over the 120b model The difference of quality and accuracy of the responses between the two is vastly different though, if tok/s isn't your biggest priority, especially when using reasoning_effort \"high\". 20B works great for small-ish text summarization and title generation, but for even moderately difficult programming tasks, 20B fails repeatedly while 120B gets it right on t", "context": "", "use_case": "programming tasks", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}, {"id": 733032305, "item_id": 45814519, "url": "https://news.ycombinator.com/item?id=45814519", "by": "embedding-shape", "date": "2025-11-04", "stance": "recommend", "quote": "The 20B is definitely worse than 120B for me in every case and scenario", "excerpt": "> But the 120b model has just as bad if not worse formatting issues, compared to the 20b one What runtime/tools are you using? Haven't been my experience at all, but I've also mostly used it via llama.cpp and my own \"coding agent\". It was slightly tricky to get the Harmony parsing in place and working correct, but once that's in place, I haven't seen any formatting issues at all? The 20B is definitely worse than 120B", "context": "", "use_case": "local coding assistant", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}]}, {"label": "Private data", "summary": "One user replaced ChatGPT with it and likes that all data stays at home.", "quotes": [{"id": 732478098, "item_id": 45779881, "url": "https://news.ycombinator.com/item?id=45779881", "by": "Greenpants", "date": "2025-11-01", "stance": "recommend", "quote": "I rarely ever use ChatGPT anymore. Love that all data stays at home.", "excerpt": "I got a personal Mac Studio M4 Max with 128GB RAM for a silent, relatively power-efficient yet powerful home server. It runs Ollama + Open WebUI with GPT-OSS 120b as well as GLM4.5-Air (default quantisations). I rarely ever use ChatGPT anymore. Love that all data stays at home. I connect remotely only via VPN (my phone enables this automatically via Tasker). I'm 50% brainstorming ideas with it, asking critical questi", "context": "", "use_case": "local brainstorming and coding", "switched_from": "ChatGPT", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}]}], "caveats": [{"label": "20B mistakes", "summary": "The 20B model is described as limited, inaccurate, or prone to errors like looping searches.", "quotes": [{"id": 769439041, "item_id": 48089940, "url": "https://news.ycombinator.com/item?id=48089940", "by": "quacker", "date": "2026-05-10", "stance": "neutral", "quote": "GPT OSS 20B was usable but slow, and actually frequently made mistakes", "excerpt": "I could have used this article before I spent the weekend arriving to the same conclusion! Same laptop, and my contrived test was having it fix 50 or so lint errors in a small vibe-coded C++ repo. I wanted it to be able to handle a bunch of small tasks without getting stuck too often. GPT OSS 20B was usable but slow, and actually frequently made mistakes like adding or duplicating statements unnecessarily, listing th", "context": "", "use_case": "local lint fixing", "switched_from": "", "story": "Running local models on an M4 with 24GB memory"}, {"id": 734912880, "item_id": 45932055, "url": "https://news.ycombinator.com/item?id=45932055", "by": "ticulatedspline", "date": "2025-11-14", "stance": "neutral", "quote": "I tried gpt-oss-20b (my go-to local) and it looks ok though not very accurate.", "excerpt": "This is cool, interesting to see how consistent some models are (both in success and failure) I tried gpt-oss-20b (my go-to local) and it looks ok though not very accurate. It decided to omit numbers. It also took 4500 tokens while thinking. I'd be interested in seeing it with some more token leeway as well as comparing two or more similar prompts. like using \"current time\" instead of \"${time}\" and being more prescri", "context": "", "use_case": "local LLM", "switched_from": "", "story": "AI World Clocks"}, {"id": 732522611, "item_id": 45782663, "url": "https://news.ycombinator.com/item?id=45782663", "by": "nickthegreek", "date": "2025-11-01", "stance": "warn", "quote": "oss20b seemed to be looping searches", "excerpt": "just setup searxng yesterday and mcp for it in lm studio to be able to search the net for answers to simple queries. small ibm granite worked surpassingly well, while oss20b seemed to be looping searches.", "context": "", "use_case": "web search queries", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}]}, {"label": "Setup quirks", "summary": "Good results can require passing back reasoning tokens, Harmony parsing, or pinning tool versions.", "quotes": [{"id": 741833552, "item_id": 46364597, "url": "https://news.ycombinator.com/item?id=46364597", "by": "nekitamo", "date": "2025-12-23", "stance": "recommend", "quote": "it started functioning extremely well, and it's the main model I use for my homemade agents.", "excerpt": "GPT-oss-120B was also completely failing for me, until someone on reddit pointed out that you need to pass back in the reasoning tokens when generating a response. One way to do this is described here: https://openrouter.ai/docs/guides/best-practices/reasoning-tokens Once I did that it started functioning extremely well, and it's the main model I use for my homemade agents. Many LLM libraries/services/frontends don't", "context": "builds own agents", "use_case": "homemade agents", "switched_from": "", "story": "GLM-4.7: Advancing the Coding Capability"}, {"id": 733032305, "item_id": 45814519, "url": "https://news.ycombinator.com/item?id=45814519", "by": "embedding-shape", "date": "2025-11-04", "stance": "recommend", "quote": "The 20B is definitely worse than 120B for me in every case and scenario", "excerpt": "> But the 120b model has just as bad if not worse formatting issues, compared to the 20b one What runtime/tools are you using? Haven't been my experience at all, but I've also mostly used it via llama.cpp and my own \"coding agent\". It was slightly tricky to get the Harmony parsing in place and working correct, but once that's in place, I haven't seen any formatting issues at all? The 20B is definitely worse than 120B", "context": "", "use_case": "local coding assistant", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}, {"id": 763951426, "item_id": 47746964, "url": "https://news.ycombinator.com/item?id=47746964", "by": "vsrinivas", "date": "2026-04-12", "stance": "neutral", "quote": "I use the same, w/ both gemma4 and gpt-oss-*", "excerpt": "Hey - I use the same, w/ both gemma4 and gpt-oss-*; some things I have to do for a good experience: 1) Pin to an earlier version of codex (sorry) - 0.55 is the best experience IME, but YMMV (see https://github.com/openai/codex/issues/11940, https://github.com/openai/codex/issues/8272). 2) Use the older completions endpoint (llama.cpp's responses support is incomplete - https://github.com/ggml-org/llama.cpp/issues/191", "context": "", "use_case": "local model with Codex", "switched_from": "", "story": "I ran Gemma 4 as a local model in Codex CLI"}]}, {"label": "Below frontier", "summary": "Commenters say it is not as good as frontier models and struggles in complex scenarios.", "quotes": [{"id": 733461746, "item_id": 45841359, "url": "https://news.ycombinator.com/item?id=45841359", "by": "ericd", "date": "2025-11-06", "stance": "neutral", "quote": "gpt-oss 120b is obviously not as good as one of the frontier models", "excerpt": "Absolutely, especially the part about just rolling your own alternative to Claude Code - build your own lightsaber. Having your coding agent improve itself is a pretty magical experience. And then you can trivially swap in whatever model you want (Cerebras is crazy fast, for example, which makes a big difference for these many-turn tool call conversations with big lumps of context, though gpt-oss 120b is obviously no", "context": "", "use_case": "agent model", "switched_from": "", "story": "You should write an agent"}, {"id": 747173937, "item_id": 46698371, "url": "https://news.ycombinator.com/item?id=46698371", "by": "lout332", "date": "2026-01-20", "stance": "compare", "quote": "GPT-OSS dominates simple 3-chip games (67% win rate) but collapses to 10% in complex 7-chip games", "excerpt": "We used \"So Long Sucker\" (1950), a 4-player negotiation/betrayal game designed by John Nash and others, as a deception benchmark for modern LLMs. The game has a brutal property: you need allies to survive, but only one player can win, so every alliance must eventually end in betrayal. We ran 162 AI vs AI games (15,736 decisions, 4,768 messages) across Gemini 3 Flash, GPT-OSS 120B, Kimi K2, and Qwen3 32B. Key findings", "context": "", "use_case": "deception benchmark", "switched_from": "", "story": "Which AI Lies Best? LLMs play a 1950s betrayal game by John Nash"}, {"id": 762190001, "item_id": 47636875, "url": "https://news.ycombinator.com/item?id=47636875", "by": "survirtual", "date": "2026-04-04", "stance": "warn", "quote": "GPT-OSS119B failed at my evals.", "excerpt": "I have been running nearly the same experiments for the same reasons. It has been a lot of tweaking and patching because I am on a Strix Halo system with amd gfx1191, but the LLM component is working nicely. The evaluation I use is having it one shot a 3d scene using threejs. After everything, the output was comparable to Claude Sonnet (Opus actually does worse on this task strangely). For my local setup, I have sett", "context": "", "use_case": "local LLM", "switched_from": "", "story": "Vitalik Buterin \u2013 \"My self-sovereign / local / private / secure LLM setup\""}]}, {"label": "120B formatting", "summary": "One user found the 120B model had formatting issues and hallucinated more than the 20B.", "quotes": [{"id": 732807296, "item_id": 45800456, "url": "https://news.ycombinator.com/item?id=45800456", "by": "gunalx", "date": "2025-11-03", "stance": "compare", "quote": "the 120b model has just as bad if not worse formatting issues, compared to the 20b one", "excerpt": "But the 120b model has just as bad if not worse formatting issues, compared to the 20b one. For simple refactorings, or chatting about possible solutions i actually feel teh 20b halucinates less than the 120b, even if it is less competent. Migth also be because of 120b not liking being in q8, or not being properly deployed.", "context": "", "use_case": "coding assistance", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}, {"id": 732807297, "item_id": 45800456, "url": "https://news.ycombinator.com/item?id=45800456", "by": "gunalx", "date": "2025-11-03", "stance": "compare", "quote": "i actually feel teh 20b halucinates less than the 120b, even if it is less competent", "excerpt": "But the 120b model has just as bad if not worse formatting issues, compared to the 20b one. For simple refactorings, or chatting about possible solutions i actually feel teh 20b halucinates less than the 120b, even if it is less competent. Migth also be because of 120b not liking being in q8, or not being properly deployed.", "context": "", "use_case": "refactoring and chatting about solutions", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}]}], "who": [{"by": "0x20Fearless", "context": "16GB VRAM and 64GB RAM", "url": "https://news.ycombinator.com/item?id=50102220", "date": "2026-10-15"}, {"by": "nyrikki", "context": "30 years of experience", "url": "https://news.ycombinator.com/item?id=49049803", "date": "2026-07-25"}, {"by": "rpdillon", "context": "runs a Strix Halo box", "url": "https://news.ycombinator.com/item?id=47614043", "date": "2026-04-02"}, {"by": "nekitamo", "context": "builds homemade agents", "url": "https://news.ycombinator.com/item?id=46881502", "date": "2026-02-03"}, {"by": "EagnaIonat", "context": "M4 Max 128GB Mac user", "url": "https://news.ycombinator.com/item?id=45820894", "date": "2025-11-05"}], "tags": ["self-hostable", "for Mac users", "for building agents", "for coding", "for privacy"], "use_cases": ["local llm", "local inference", "local llm inference", "homemade agents", "local coding assistant", "local llm for small tasks"], "compared": [{"slug": "qwen-3-5", "name": "Qwen 3.5", "co": 7, "switched_from": 0, "switched_to": 1}, {"slug": "llama-cpp", "name": "llama.cpp", "co": 9, "switched_from": 0, "switched_to": 0}, {"slug": "qwen-3-6", "name": "Qwen 3.6", "co": 5, "switched_from": 0, "switched_to": 0}, {"slug": "gemma", "name": "Gemma", "co": 5, "switched_from": 0, "switched_to": 0}, {"slug": "ollama", "name": "Ollama", "co": 5, "switched_from": 0, "switched_to": 0}, {"slug": "lm-studio", "name": "LM Studio", "co": 4, "switched_from": 0, "switched_to": 0}, {"slug": "chatgpt", "name": "ChatGPT", "co": 2, "switched_from": 1, "switched_to": 0}, {"slug": "home-assistant", "name": "Home Assistant", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=45771870", "title": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop", "mentions": 10}, {"url": "https://news.ycombinator.com/item?id=50242939", "title": "Anthropic S-1", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=50097369", "title": "LFM2.5-24B-A2B: Built for fast on-device agents", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=50042333", "title": "Show HN: BetterWispr \u2013 Free, open-source dictation for Mac", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49508982", "title": "Apple caught off guard by AI demand for Mac Mini and Mac Studio", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49433316", "title": "New Mac Studio with M5 Max and M5 Ultra", "mentions": 1}], "sentiment_by_month": {"2025-11": {"recommend": 3, "warn": 1, "compare": 1, "neutral": 2}, "2025-12": {"recommend": 2}, "2026-01": {"neutral": 1, "recommend": 2, "compare": 1}, "2026-02": {"recommend": 2}, "2026-03": {}, "2026-04": {"neutral": 2, "recommend": 1, "warn": 1}, "2026-05": {"neutral": 2, "recommend": 1}, "2026-06": {"recommend": 2, "compare": 1, "neutral": 1}, "2026-07": {"neutral": 2}, "2026-08": {"recommend": 1, "neutral": 2}, "2026-09": {}, "2026-10": {"recommend": 2, "neutral": 1}}, "quotes": [{"id": 804880544, "item_id": 50305034, "url": "https://news.ycombinator.com/item?id=50305034", "by": "Breza", "date": "2026-10-30", "stance": "recommend", "quote": "I find myself firing up GPT-0SS-20b on my laptop for smaller tasks", "excerpt": "I agree with you. I use LLMs for a lot of different things. But increasingly, I find myself firing up GPT-0SS-20b on my laptop for smaller tasks. And I know there are better models I just haven't tried yet.", "context": "", "use_case": "local LLM for small tasks", "switched_from": "", "story": "Anthropic S-1"}, {"id": 801635520, "item_id": 50102220, "url": "https://news.ycombinator.com/item?id=50102220", "by": "0x20Fearless", "date": "2026-10-15", "stance": "neutral", "quote": "I do run the gpt-oss-20B fast on my 16GB vram, but it\u2019s, limited.", "excerpt": "I love your detailed explanation of things that I had little to no knowledge of, but I\u2019m also stuck wondering if I\u2019m using my setup correctly, as I have 16 GB of VRAM available but also 64 GB of regular RAM in my computer. Every time I use some setup or some configuration that overlaps or overflows between the two RAMs, my token per second drops significantly, me thinking that it is not set up correctly, so I\u2019m stuck", "context": "16GB VRAM and 64GB RAM", "use_case": "local LLM on 16GB VRAM", "switched_from": "", "story": "LFM2.5-24B-A2B: Built for fast on-device agents"}, {"id": 800780321, "item_id": 50048770, "url": "https://news.ycombinator.com/item?id=50048770", "by": "zalthor", "date": "2026-10-11", "stance": "recommend", "quote": "I've now been playing around with gpt-oss-20b on Groq, which appears to be fast and effective.", "excerpt": "Interestingly enough, I was working on something very similar this past few of days[1]! A couple of things that I did differently - One, in the post-transcription cleanup phase, instead of programmatically fixing the text, I just feed the text with a prompt + dictionary of words into an LLM. Initially tried doing this with the on-device model from Apple in macOS 27, but that was too slow. I've now been playing around", "context": "", "use_case": "dictation text cleanup", "switched_from": "", "story": "Show HN: BetterWispr \u2013 Free, open-source dictation for Mac"}, {"id": 792161105, "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": 790939041, "item_id": 49433690, "url": "https://news.ycombinator.com/item?id=49433690", "by": "super_mario", "date": "2026-08-25", "stance": "neutral", "quote": "I can run local LLMs like Gemma 3/4, gpt-oss 120b all day long", "excerpt": "I was in the same situation, I used maxed out 15'' M3 Max MacBook Pro docked to Studio Display closed on vertical stand behind the screen. It was fine for office work, but running local LLMs would definitely overheat it. The battery started degrading purely due to heat issues. And it was audible as well. I decided to get Mac Studio M4 Max, also all maxed out config and the cooling is so much better that I can run loc", "context": "", "use_case": "local LLM inference", "switched_from": "", "story": "New Mac Studio with M5 Max and M5 Ultra"}, {"id": 787288389, "item_id": 49205524, "url": "https://news.ycombinator.com/item?id=49205524", "by": "alasr", "date": "2026-08-06", "stance": "recommend", "quote": "start with Qwen3.6, Google's Gemma4, OpenAI's gpt-oss-20b or Nvidia's Nemotron", "excerpt": "- See \"Llama.cpp vs Ollama\"[0] and \"Llama.cpp vs LM Studio\"[1] for a high-level features comparison and find out which one suite you best (check out the \"Target Users\" category[0][1], at least). - If you're still not sure, \"LM Studio\"[2] is not bad to start with as you'll be able to download, start/stop/manage and chat with your LLM model all in one place: a single desktop app. Also, once installed, enable \"Developer", "context": "", "use_case": "local LLM", "switched_from": "", "story": "Qwen3.8-Max: A New Bar for Coding and Cowork"}, {"id": 784796851, "item_id": 49049803, "url": "https://news.ycombinator.com/item?id=49049803", "by": "nyrikki", "date": "2026-07-25", "stance": "neutral", "quote": "GPT-OSS 120b on pure cpu (slow)", "excerpt": "I don\u2019t know if others would find this useful, but previous did have custom harnesses etc.. but tools have improved so much that I drastically simplified. That said, even the foundational models fail at the hard parts of my code so I use it opportunistically. I have reduced down to just using zed, will three locally hosted models. Qwen 3.6 27b on 1x3090 llama.cpp with 128k context ~50tps Qwen 3.6 35B-A3B on 1x titan ", "context": "30 years of experience", "use_case": "local coding", "switched_from": "", "story": "Open-weight AI is having its Kubernetes moment. Let's not ruin it"}, {"id": 782179569, "item_id": 48886223, "url": "https://news.ycombinator.com/item?id=48886223", "by": "7402", "date": "2026-07-12", "stance": "neutral", "quote": "gpt-oss-120b-F16.gguf", "excerpt": "With llama.cpp, Intel i9-13900KS CPU, 96 GB RAM, RTX 4070 running locally. The models I'm using right now with that are: gpt-oss-120b-F16.gguf Qwen_Qwen3.5-27B-Q4_K_M.gguf Qwen3.6-35B-A3B-UD-Q5_K_XL.gguf gemma-4-31B-it-UD-Q6_K_XL.gguf", "context": "", "use_case": "local LLM", "switched_from": "", "story": "Ask HN: How do you use LLMs for private discussions?"}, {"id": 777055426, "item_id": 48565964, "url": "https://news.ycombinator.com/item?id=48565964", "by": "EagnaIonat", "date": "2026-06-16", "stance": "neutral", "quote": "gpt-oss:120b  (less so now)", "excerpt": "> thers, like coding or more advanced agentic workflows can demand much more powerful models. You can do coding and agentic fine. For coding I use qwen3.6:35b-mlx and agentic granite4.1:3b works fine. These are the models I use. - granite4.1:3b - granite4.1:30b - gpt-oss:20b - gpt-oss:120b (less so now) - mistral-small3.2 - qwen3.6:35b-mlx There will always be use cases that don't sit on your laptop, but most of what", "context": "", "use_case": "local inference", "switched_from": "", "story": null}, {"id": 776434640, "item_id": 48527165, "url": "https://news.ycombinator.com/item?id=48527165", "by": "pixel_popping", "date": "2026-06-14", "stance": "compare", "quote": "I don't see how GPT OSS is an \"alternative\" to Claude, because both are LLMs?", "excerpt": "I don't see how GPT OSS is an \"alternative\" to Claude, because both are LLMs? that's like comparing making an alternative to a 4K TV with a gameboy :/", "context": "", "use_case": "", "switched_from": "", "story": "Build Claude Alternative in Cloud in 20mins"}, {"id": 775981172, "item_id": 48498823, "url": "https://news.ycombinator.com/item?id=48498823", "by": "inventor7777", "date": "2026-06-11", "stance": "recommend", "quote": "runs GPT-OSS 20B at over 100 tok/sec", "excerpt": "5-10 sessions?! Wow, not sure how you keep track of those. For all my daily use + Codex Desktop and some local AIs in Ollama/LM Studio (which power my Home Assistant Voice Assistants) I have an Mac Studio M4 Max with 48GB of RAM (Which I bought early last year IIRC) It's a complete beast and runs GPT-OSS 20B at over 100 tok/sec. The GUI never slows down, and I run Sequoia still, so it is always smooth and polished. I", "context": "", "use_case": "local inference", "switched_from": "", "story": "Ask HN: What computer are you using for AI coding tools?"}, {"id": 775033664, "item_id": 48439604, "url": "https://news.ycombinator.com/item?id=48439604", "by": "computerex", "date": "2026-06-07", "stance": "recommend", "quote": "It\u2019s highly capable in agentic settings, people use it for coding all the time.", "excerpt": "You are actually referring to open weight models, not open source. Gpt-OSS is an example of an open weight model. It\u2019s highly capable in agentic settings, people use it for coding all the time. My greater point remains. Models like the qwen variants, minimax, k2.5, glm models are available by American providers like AWS at a much cheaper price than api offerings from the big three LLM providers. Your point about Chin", "context": "", "use_case": "agentic coding", "switched_from": "", "story": "The OnlyFans Economy of American AI"}, {"id": 770750021, "item_id": 48171876, "url": "https://news.ycombinator.com/item?id=48171876", "by": "zambelli", "date": "2026-05-17", "stance": "neutral", "quote": "GPT-OSS-120B for self-hosted coding or more reasoning-intensive agentic flows.", "excerpt": "I use Opus 4.7 for personal stuff (basically for everything), but have been considering gpt-5.5 given all I hear about it. At work I use 4.6 because we don't have 4.7 yet...zzz... I also do a LOT of personal/portfolio work with self-hosted models. Ministral-3-14B-Reasoning for validating concepts, MVPs, etc and some prod systems (punches above its weight class). Qwen3.6-35B-A3B for self-hosted coding (custom harness)", "context": "", "use_case": "self-hosted coding, agentic flows", "switched_from": "", "story": "Ask HN: What LLM models are you using and why?"}, {"id": 769517762, "item_id": 48094860, "url": "https://news.ycombinator.com/item?id=48094860", "by": "tristor", "date": "2026-05-11", "stance": "recommend", "quote": "Qwen 3.5, Gemma 4, even gpt-oss-120b are solid at reasonable quants if they have the right data.", "excerpt": "The biggest challenge I have with local models right now (and I use them extensively) is search integration and tool calling. The thing that Claude and ChatGPT get right for most general purpose use cases which is hard to do with a local model is the model deciding when to search vs use its built-in training, and having strong search tooling, as well as tool calling for additional data sources via MCP. If you can inc", "context": "", "use_case": "local general-purpose use", "switched_from": "", "story": "Local AI needs to be the norm"}, {"id": 769439041, "item_id": 48089940, "url": "https://news.ycombinator.com/item?id=48089940", "by": "quacker", "date": "2026-05-10", "stance": "neutral", "quote": "GPT OSS 20B was usable but slow, and actually frequently made mistakes", "excerpt": "I could have used this article before I spent the weekend arriving to the same conclusion! Same laptop, and my contrived test was having it fix 50 or so lint errors in a small vibe-coded C++ repo. I wanted it to be able to handle a bunch of small tasks without getting stuck too often. GPT OSS 20B was usable but slow, and actually frequently made mistakes like adding or duplicating statements unnecessarily, listing th", "context": "", "use_case": "local lint fixing", "switched_from": "", "story": "Running local models on an M4 with 24GB memory"}, {"id": 763951426, "item_id": 47746964, "url": "https://news.ycombinator.com/item?id=47746964", "by": "vsrinivas", "date": "2026-04-12", "stance": "neutral", "quote": "I use the same, w/ both gemma4 and gpt-oss-*", "excerpt": "Hey - I use the same, w/ both gemma4 and gpt-oss-*; some things I have to do for a good experience: 1) Pin to an earlier version of codex (sorry) - 0.55 is the best experience IME, but YMMV (see https://github.com/openai/codex/issues/11940, https://github.com/openai/codex/issues/8272). 2) Use the older completions endpoint (llama.cpp's responses support is incomplete - https://github.com/ggml-org/llama.cpp/issues/191", "context": "", "use_case": "local model with Codex", "switched_from": "", "story": "I ran Gemma 4 as a local model in Codex CLI"}, {"id": 762190001, "item_id": 47636875, "url": "https://news.ycombinator.com/item?id=47636875", "by": "survirtual", "date": "2026-04-04", "stance": "warn", "quote": "GPT-OSS119B failed at my evals.", "excerpt": "I have been running nearly the same experiments for the same reasons. It has been a lot of tweaking and patching because I am on a Strix Halo system with amd gfx1191, but the LLM component is working nicely. The evaluation I use is having it one shot a 3d scene using threejs. After everything, the output was comparable to Claude Sonnet (Opus actually does worse on this task strangely). For my local setup, I have sett", "context": "", "use_case": "local LLM", "switched_from": "", "story": "Vitalik Buterin \u2013 \"My self-sovereign / local / private / secure LLM setup\""}, {"id": 761924928, "item_id": 47620308, "url": "https://news.ycombinator.com/item?id=47620308", "by": "evilduck", "date": "2026-04-02", "stance": "recommend", "quote": "gpt-oss-120b is over 600 tokens/s PP for all but one backend.", "excerpt": "In terms of ability, maybe, in terms of speed, it's not even close. Check out the Prompt Processing speeds between them: https://kyuz0.github.io/amd-strix-halo-toolboxes/ gpt-oss-120b is over 600 tokens/s PP for all but one backend. nemotron-3-super is at best 260 tokens/s PP. Comparing token generation, it's again like 50 tokens/sec vs 15 tokens/sec That really bogs down agentic tooling. Something needs to be catego", "context": "", "use_case": "local agentic inference", "switched_from": "", "story": "Google releases Gemma 4 open models"}, {"id": 761824690, "item_id": 47614043, "url": "https://news.ycombinator.com/item?id=47614043", "by": "rpdillon", "date": "2026-04-02", "stance": "neutral", "quote": "I get about 50 tokens a second with GPT OSS 120B.", "excerpt": "Been running lemonade for some time on my Strix Halo box. It dispatches out to other backends that they include, like diffusion and llama. I actually don't like their combined server, and what I use instead is their llama CPP build for ROCm. https://github.com/lemonade-sdk/llamacpp-rocm But I'm not doing anything with images or audio. I get about 50 tokens a second with GPT OSS 120B. As others have pointed out, the N", "context": "runs a Strix Halo box", "use_case": "local inference", "switched_from": "", "story": "Lemonade by AMD: a fast and open source local LLM server using GPU and NPU"}, {"id": 755262544, "item_id": 47203909, "url": "https://news.ycombinator.com/item?id=47203909", "by": "carlio", "date": "2026-02-28", "stance": "recommend", "quote": "it can run a 120b model (gpt-oss:120b) just fine.", "excerpt": "An AMD AI max+ 395 - I use the one from frame.work (https://frame.work/de/en/desktop) with 128GB unified RAM and it can run a 120b model (gpt-oss:120b) just fine. See Wendel's review here - https://www.youtube.com/watch?v=L-xgMQ-7lW0 There are other mini-pc manufacturers, the mainboard is the important part.", "context": "", "use_case": "local LLM inference", "switched_from": "", "story": "Qwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers"}, {"id": 750104032, "item_id": 46881502, "url": "https://news.ycombinator.com/item?id=46881502", "by": "nekitamo", "date": "2026-02-03", "stance": "recommend", "quote": "Once I did that it started functioning extremely well, and it's the main model I use for my homemade agents.", "excerpt": "This is a common problem for people trying to run the GPT-oss models themselves. Reposting my comment here: GPT-oss-120B was also completely failing for me, until someone on reddit pointed out that you need to pass back in the reasoning tokens when generating a response. One way to do this is described here: https://openrouter.ai/docs/guides/best-practices/reasoning-tokens Once I did that it started functioning extre", "context": "builds homemade agents", "use_case": "homemade agents", "switched_from": "", "story": "Qwen3-Coder-Next"}, {"id": 747807504, "item_id": 46737969, "url": "https://news.ycombinator.com/item?id=46737969", "by": "khimaros", "date": "2026-01-23", "stance": "recommend", "quote": "i used this with a self-hosted gpt-oss-120b, which turned out very well.", "excerpt": "there are two generation modes: 1) a basic single voice narrator and 2) a \"dramatization\" mode which generates a cast list from the book contents and converts the entire text into a script with distinct voices and instructions for each segment. i used this with a self-hosted gpt-oss-120b, which turned out very well. the resulting mp3 files work well with the open source \"Voice\" audiobook player on F-Droid. feedback w", "context": "", "use_case": "generating audiobook scripts", "switched_from": "", "story": "Convert ePub files to audiobooks using qwen3-TTS"}, {"id": 747173937, "item_id": 46698371, "url": "https://news.ycombinator.com/item?id=46698371", "by": "lout332", "date": "2026-01-20", "stance": "compare", "quote": "GPT-OSS dominates simple 3-chip games (67% win rate) but collapses to 10% in complex 7-chip games", "excerpt": "We used \"So Long Sucker\" (1950), a 4-player negotiation/betrayal game designed by John Nash and others, as a deception benchmark for modern LLMs. The game has a brutal property: you need allies to survive, but only one player can win, so every alliance must eventually end in betrayal. We ran 162 AI vs AI games (15,736 decisions, 4,768 messages) across Gemini 3 Flash, GPT-OSS 120B, Kimi K2, and Qwen3 32B. Key findings", "context": "", "use_case": "deception benchmark", "switched_from": "", "story": "Which AI Lies Best? LLMs play a 1950s betrayal game by John Nash"}, {"id": 745405825, "item_id": 46587864, "url": "https://news.ycombinator.com/item?id=46587864", "by": "river_otter", "date": "2026-01-12", "stance": "recommend", "quote": "I've been using gpt-oss-20b which would fit on your GPU and I've found useful for basic tasks", "excerpt": "makes sense. You could look at something like https://github.com/musistudio/claude-code-router if at some point you're interested in going down that path. I've been using gpt-oss-20b which would fit on your GPU and I've found useful for basic tasks like recipe creation and agentic tool usage (I use it with Notion MCP tools)", "context": "", "use_case": "recipes and agentic tool usage", "switched_from": "", "story": "2026 is the Year of Self-hosting"}, {"id": 744837376, "item_id": 46552336, "url": "https://news.ycombinator.com/item?id=46552336", "by": "rdos", "date": "2026-01-09", "stance": "neutral", "quote": "I tested this locally and got the same result with gpt-oss 120b.", "excerpt": "This is very interesting. Especially the last part where it shows gpt-5.2 and gpt-oss and their very similar and unique outcome of being 90%+ Serious. I tested this locally and got the same result with gpt-oss 120b. But only on the default 'medium' reasoning effort. When I used 'low' I kept getting more playful responses with emojis and when I used 'high' I kept getting more guessing responses. I had a lot of fun wit", "context": "", "use_case": "testing LLM personality locally", "switched_from": "", "story": "Task-free intelligence testing of LLMs"}, {"id": 742139857, "item_id": 46383741, "url": "https://news.ycombinator.com/item?id=46383741", "by": "sgk284", "date": "2025-12-25", "stance": "recommend", "quote": "We use Groq with gpt-oss-20b, which gives great results and only adds ~250ms to the processing pipeline.", "excerpt": "Reranking is definitely the way to go. We personally found common reranker models to be a little too opaque (can't explain to the user why this result was picked) and not quite steerable enough, so we just use another LLM for reranking. We open-sourced our impl just this week: https://github.com/with-logic/intent We use Groq with gpt-oss-20b, which gives great results and only adds ~250ms to the processing pipeline. ", "context": "", "use_case": "reranking search results", "switched_from": "", "story": "Using Vectorize to build an unreasonably good search engine in 160 lines of code"}, {"id": 741833552, "item_id": 46364597, "url": "https://news.ycombinator.com/item?id=46364597", "by": "nekitamo", "date": "2025-12-23", "stance": "recommend", "quote": "it started functioning extremely well, and it's the main model I use for my homemade agents.", "excerpt": "GPT-oss-120B was also completely failing for me, until someone on reddit pointed out that you need to pass back in the reasoning tokens when generating a response. One way to do this is described here: https://openrouter.ai/docs/guides/best-practices/reasoning-tokens Once I did that it started functioning extremely well, and it's the main model I use for my homemade agents. Many LLM libraries/services/frontends don't", "context": "builds own agents", "use_case": "homemade agents", "switched_from": "", "story": "GLM-4.7: Advancing the Coding Capability"}, {"id": 734912880, "item_id": 45932055, "url": "https://news.ycombinator.com/item?id=45932055", "by": "ticulatedspline", "date": "2025-11-14", "stance": "neutral", "quote": "I tried gpt-oss-20b (my go-to local) and it looks ok though not very accurate.", "excerpt": "This is cool, interesting to see how consistent some models are (both in success and failure) I tried gpt-oss-20b (my go-to local) and it looks ok though not very accurate. It decided to omit numbers. It also took 4500 tokens while thinking. I'd be interested in seeing it with some more token leeway as well as comparing two or more similar prompts. like using \"current time\" instead of \"${time}\" and being more prescri", "context": "", "use_case": "local LLM", "switched_from": "", "story": "AI World Clocks"}, {"id": 733461746, "item_id": 45841359, "url": "https://news.ycombinator.com/item?id=45841359", "by": "ericd", "date": "2025-11-06", "stance": "neutral", "quote": "gpt-oss 120b is obviously not as good as one of the frontier models", "excerpt": "Absolutely, especially the part about just rolling your own alternative to Claude Code - build your own lightsaber. Having your coding agent improve itself is a pretty magical experience. And then you can trivially swap in whatever model you want (Cerebras is crazy fast, for example, which makes a big difference for these many-turn tool call conversations with big lumps of context, though gpt-oss 120b is obviously no", "context": "", "use_case": "agent model", "switched_from": "", "story": "You should write an agent"}, {"id": 733134305, "item_id": 45820894, "url": "https://news.ycombinator.com/item?id=45820894", "by": "EagnaIonat", "date": "2025-11-05", "stance": "recommend", "quote": "gpt-oss. 20b or 120b depending on complexity of use cases.", "excerpt": "The more recent LLMs work fine on an M1 mac. Can't speak for Windows/Linux. There was even a recent release of Granite4 that runs on a Raspberry Pi. https://github.com/Jewelzufo/granitepi-4-nano For my local work I use Ollama. (M4 Max 128GB) - gpt-oss. 20b or 120b depending on complexity of use cases. - granite4 for speed and lower complexity (around the same as gpt20b).", "context": "M4 Max 128GB Mac user", "use_case": "local LLM work", "switched_from": "", "story": "Developers are choosing older AI models"}, {"id": 733032305, "item_id": 45814519, "url": "https://news.ycombinator.com/item?id=45814519", "by": "embedding-shape", "date": "2025-11-04", "stance": "recommend", "quote": "The 20B is definitely worse than 120B for me in every case and scenario", "excerpt": "> But the 120b model has just as bad if not worse formatting issues, compared to the 20b one What runtime/tools are you using? Haven't been my experience at all, but I've also mostly used it via llama.cpp and my own \"coding agent\". It was slightly tricky to get the Harmony parsing in place and working correct, but once that's in place, I haven't seen any formatting issues at all? The 20B is definitely worse than 120B", "context": "", "use_case": "local coding assistant", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}, {"id": 733032306, "item_id": 45814519, "url": "https://news.ycombinator.com/item?id=45814519", "by": "embedding-shape", "date": "2025-11-04", "stance": "compare", "quote": "The 20B is definitely worse than 120B for me in every case and scenario, but it is a lot faster.", "excerpt": "> But the 120b model has just as bad if not worse formatting issues, compared to the 20b one What runtime/tools are you using? Haven't been my experience at all, but I've also mostly used it via llama.cpp and my own \"coding agent\". It was slightly tricky to get the Harmony parsing in place and working correct, but once that's in place, I haven't seen any formatting issues at all? The 20B is definitely worse than 120B", "context": "", "use_case": "local coding assistant", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}, {"id": 732807296, "item_id": 45800456, "url": "https://news.ycombinator.com/item?id=45800456", "by": "gunalx", "date": "2025-11-03", "stance": "compare", "quote": "the 120b model has just as bad if not worse formatting issues, compared to the 20b one", "excerpt": "But the 120b model has just as bad if not worse formatting issues, compared to the 20b one. For simple refactorings, or chatting about possible solutions i actually feel teh 20b halucinates less than the 120b, even if it is less competent. Migth also be because of 120b not liking being in q8, or not being properly deployed.", "context": "", "use_case": "coding assistance", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}, {"id": 732807297, "item_id": 45800456, "url": "https://news.ycombinator.com/item?id=45800456", "by": "gunalx", "date": "2025-11-03", "stance": "compare", "quote": "i actually feel teh 20b halucinates less than the 120b, even if it is less competent", "excerpt": "But the 120b model has just as bad if not worse formatting issues, compared to the 20b one. For simple refactorings, or chatting about possible solutions i actually feel teh 20b halucinates less than the 120b, even if it is less competent. Migth also be because of 120b not liking being in q8, or not being properly deployed.", "context": "", "use_case": "refactoring and chatting about solutions", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}, {"id": 732548176, "item_id": 45784261, "url": "https://news.ycombinator.com/item?id=45784261", "by": "Greenpants", "date": "2025-11-01", "stance": "recommend", "quote": "I've read that GPT-OSS:20b is still a very powerful model", "excerpt": "The TODOs really help me get my logic sorted out first in pseudocode. Glad to inspire someone else with it! I've read that GPT-OSS:20b is still a very powerful model, I believe it fits in your Mac's RAM as well and could still be quite fast to output. For me personally, only the more complex questions require a better model than local ones. And then I'm often wondering if LLMs are the right tool to solve the complexi", "context": "", "use_case": "local model on 32GB Mac", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}, {"id": 732522611, "item_id": 45782663, "url": "https://news.ycombinator.com/item?id=45782663", "by": "nickthegreek", "date": "2025-11-01", "stance": "warn", "quote": "oss20b seemed to be looping searches", "excerpt": "just setup searxng yesterday and mcp for it in lm studio to be able to search the net for answers to simple queries. small ibm granite worked surpassingly well, while oss20b seemed to be looping searches.", "context": "", "use_case": "web search queries", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}, {"id": 732500112, "item_id": 45781257, "url": "https://news.ycombinator.com/item?id=45781257", "by": "embedding-shape", "date": "2025-11-01", "stance": "recommend", "quote": "for even moderately difficult programming tasks, 20B fails repeatedly while 120B gets it right on the first try", "excerpt": "> had better results with the 20b model, over the 120b model The difference of quality and accuracy of the responses between the two is vastly different though, if tok/s isn't your biggest priority, especially when using reasoning_effort \"high\". 20B works great for small-ish text summarization and title generation, but for even moderately difficult programming tasks, 20B fails repeatedly while 120B gets it right on t", "context": "", "use_case": "programming tasks", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}, {"id": 732500113, "item_id": 45781257, "url": "https://news.ycombinator.com/item?id=45781257", "by": "embedding-shape", "date": "2025-11-01", "stance": "neutral", "quote": "20B works great for small-ish text summarization and title generation", "excerpt": "> had better results with the 20b model, over the 120b model The difference of quality and accuracy of the responses between the two is vastly different though, if tok/s isn't your biggest priority, especially when using reasoning_effort \"high\". 20B works great for small-ish text summarization and title generation, but for even moderately difficult programming tasks, 20B fails repeatedly while 120B gets it right on t", "context": "", "use_case": "summarization and title generation", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}, {"id": 732496468, "item_id": 45781029, "url": "https://news.ycombinator.com/item?id=45781029", "by": "embedding-shape", "date": "2025-11-01", "stance": "recommend", "quote": "frequent user of GPT-OSS", "excerpt": "Not parent, but frequent user of GPT-OSS, tried all different ways of running it. Choice goes something like this: - Need batching + highest total throughoutput? vLLM, complicated to deploy and install though, need special versions for top performance with GPT-OSS - Easiest to manage + fast enough: llama.cpp, easier to deploy as well (just a binary) and super fast, getting ~260 tok/s on a RTX Pro 6000 for the 20B ver", "context": "", "use_case": "local LLM", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}, {"id": 732478098, "item_id": 45779881, "url": "https://news.ycombinator.com/item?id=45779881", "by": "Greenpants", "date": "2025-11-01", "stance": "recommend", "quote": "I rarely ever use ChatGPT anymore. Love that all data stays at home.", "excerpt": "I got a personal Mac Studio M4 Max with 128GB RAM for a silent, relatively power-efficient yet powerful home server. It runs Ollama + Open WebUI with GPT-OSS 120b as well as GLM4.5-Air (default quantisations). I rarely ever use ChatGPT anymore. Love that all data stays at home. I connect remotely only via VPN (my phone enables this automatically via Tasker). I'm 50% brainstorming ideas with it, asking critical questi", "context": "", "use_case": "local brainstorming and coding", "switched_from": "ChatGPT", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}]}