GPT-OSS
OpenAI's open-weight large language models, released in 20B and 120B parameter sizes.
HN commenters like it for running locally on laptops and home servers,1234 for fast output,567 and for agentic and coding work.8910 The most common complaint is that the 20B model makes frequent mistakes on harder tasks,111213 and some say the 120B model only worked well after setup fixes like passing back reasoning tokens.141516
Why people recommend it
Runs locally · 3
Commenters run it on laptops, Macs and mini PCs, with the 120B model running fine on 128GB machines.
“I find myself firing up GPT-0SS-20b on my laptop for smaller tasks”
“it can run a 120b model (gpt-oss:120b) just fine.”
Show 1 more
“I can run local LLMs like Gemma 3/4, gpt-oss 120b all day long”
Speed · 3
Users report fast token generation locally and low latency on hosted providers like Groq.
“runs GPT-OSS 20B at over 100 tok/sec”
“gpt-oss-120b is over 600 tokens/s PP for all but one backend.”
Show 1 more
“We use Groq with gpt-oss-20b, which gives great results and only adds ~250ms to the processing pipeline.”
Agents and coding · 3
Used as the main model for homemade agents and for self-hosted coding and tool use.
“it started functioning extremely well, and it's the main model I use for my homemade agents.”
“It’s highly capable in agentic settings, people use it for coding all the time.”
Show 1 more
“GPT-OSS-120B for self-hosted coding or more reasoning-intensive agentic flows.”
120B quality · 2
The 120B model gets moderately difficult programming tasks right where the 20B fails.
“for even moderately difficult programming tasks, 20B fails repeatedly while 120B gets it right on the first try”
“The 20B is definitely worse than 120B for me in every case and scenario”
Private data · 1
One user replaced ChatGPT with it and likes that all data stays at home.
“I rarely ever use ChatGPT anymore. Love that all data stays at home.”
Caveats and dissent
20B mistakes · 3
The 20B model is described as limited, inaccurate, or prone to errors like looping searches.
“GPT OSS 20B was usable but slow, and actually frequently made mistakes”
“I tried gpt-oss-20b (my go-to local) and it looks ok though not very accurate.”
Show 1 more
“oss20b seemed to be looping searches”
Setup quirks · 3
Good results can require passing back reasoning tokens, Harmony parsing, or pinning tool versions.
“it started functioning extremely well, and it's the main model I use for my homemade agents.”
“The 20B is definitely worse than 120B for me in every case and scenario”
Show 1 more
“I use the same, w/ both gemma4 and gpt-oss-*”
Below frontier · 3
Commenters say it is not as good as frontier models and struggles in complex scenarios.
“gpt-oss 120b is obviously not as good as one of the frontier models”
“GPT-OSS dominates simple 3-chip games (67% win rate) but collapses to 10% in complex 7-chip games”
Show 1 more
“GPT-OSS119B failed at my evals.”
120B formatting · 2
One user found the 120B model had formatting issues and hallucinated more than the 20B.
“the 120b model has just as bad if not worse formatting issues, compared to the 20b one”
“i actually feel teh 20b halucinates less than the 120b, even if it is less competent”
Often compared with
1 moved from GPT-OSS to Qwen 3.5
1 moved from ChatGPT to GPT-OSS
Who recommends it in their own words
Threads where it came up
- Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop 10 mentions
- Anthropic S-1 1 mention
- LFM2.5-24B-A2B: Built for fast on-device agents 1 mention
- Show HN: BetterWispr – Free, open-source dictation for Mac 1 mention
- Apple caught off guard by AI demand for Mac Mini and Mac Studio 1 mention
- New Mac Studio with M5 Max and M5 Ultra 1 mention
Also written as gpt-oss-120b, GPT-OSS, gpt-oss, GPT-OSS 20B. Data: gpt-oss.json. Quotes are verbatim; the verdict and theme labels are written by a model from these comments, and every clause links to its sources.
All comments 40
“I find myself firing up GPT-0SS-20b on my laptop for smaller tasks”
“I do run the gpt-oss-20B fast on my 16GB vram, but it’s, limited.”
“I've now been playing around with gpt-oss-20b on Groq, which appears to be fast and effective.”
“I use Qwen/GPT-OSS on my Mac Studio via llama.cpp server.”
“I can run local LLMs like Gemma 3/4, gpt-oss 120b all day long”
“start with Qwen3.6, Google's Gemma4, OpenAI's gpt-oss-20b or Nvidia's Nemotron”
“GPT-OSS 120b on pure cpu (slow)”
“gpt-oss-120b-F16.gguf”
“gpt-oss:120b (less so now)”
“I don't see how GPT OSS is an "alternative" to Claude, because both are LLMs?”
“runs GPT-OSS 20B at over 100 tok/sec”
“It’s highly capable in agentic settings, people use it for coding all the time.”
“GPT-OSS-120B for self-hosted coding or more reasoning-intensive agentic flows.”
“Qwen 3.5, Gemma 4, even gpt-oss-120b are solid at reasonable quants if they have the right data.”
“GPT OSS 20B was usable but slow, and actually frequently made mistakes”
“I use the same, w/ both gemma4 and gpt-oss-*”
“GPT-OSS119B failed at my evals.”
“gpt-oss-120b is over 600 tokens/s PP for all but one backend.”
“I get about 50 tokens a second with GPT OSS 120B.”
“it can run a 120b model (gpt-oss:120b) just fine.”
“Once I did that it started functioning extremely well, and it's the main model I use for my homemade agents.”
“i used this with a self-hosted gpt-oss-120b, which turned out very well.”
“GPT-OSS dominates simple 3-chip games (67% win rate) but collapses to 10% in complex 7-chip games”
“I've been using gpt-oss-20b which would fit on your GPU and I've found useful for basic tasks”
“I tested this locally and got the same result with gpt-oss 120b.”
“We use Groq with gpt-oss-20b, which gives great results and only adds ~250ms to the processing pipeline.”
“it started functioning extremely well, and it's the main model I use for my homemade agents.”
“I tried gpt-oss-20b (my go-to local) and it looks ok though not very accurate.”
“gpt-oss 120b is obviously not as good as one of the frontier models”
“gpt-oss. 20b or 120b depending on complexity of use cases.”
“The 20B is definitely worse than 120B for me in every case and scenario”
“The 20B is definitely worse than 120B for me in every case and scenario, but it is a lot faster.”
“the 120b model has just as bad if not worse formatting issues, compared to the 20b one”
“i actually feel teh 20b halucinates less than the 120b, even if it is less competent”
“I've read that GPT-OSS:20b is still a very powerful model”
“oss20b seemed to be looping searches”
“for even moderately difficult programming tasks, 20B fails repeatedly while 120B gets it right on the first try”
“20B works great for small-ish text summarization and title generation”
“frequent user of GPT-OSS”
“I rarely ever use ChatGPT anymore. Love that all data stays at home.”