- Start
- Aug 5, 202595% CONFIDENCEfrom help.openai.com
OpenAI Releases Open-Weight gpt-oss-120b and gpt-oss-20b
- On 5 August 2025 OpenAI released two open-weight reasoning models, gpt-oss-120b (117 billion parameters) and gpt-oss-20b (21 billion)[1][3]
Notable features
- They are OpenAI's first open-weight language models since GPT-2: mixture-of-experts Transformers in which gpt-oss-120b activates 5.1 billion of its 117 billion parameters per token (36 layers, 128 experts) and gpt-oss-20b 3.6 billion of 21 billion (24 layers, 32 experts), with native context lengths of up to 128k tokens[4]
- OpenAI says they are comparable to o4-mini and o3-mini respectively, and gpt-oss-20b is small enough to run on a device with over 16 GB of RAM[3]
- In OpenAI's evaluations gpt-oss-120b matches or exceeds o4-mini on Codeforces, MMLU, HLE and TauBench tool calling, and does better on HealthBench and AIME 2024 and 2025[4]
- The weights are on Hugging Face, natively quantized in MXFP4 so gpt-oss-120b runs within 80 GB of memory; the models are text-only and support low, medium and high reasoning effort, full chain-of-thought, function calling, web browsing, Python code execution, Structured Outputs and fine-tuning[2][4]
- They are released under the Apache 2.0 licence for commercial and non-commercial use[3]
References 488% CONFIDENCE
The first entry is always the pin's source. Overall confidence is a weighted average of how firmly each reference supports the start and end times used above; a reference counts half as much for every 180 days older than the newest.
- [1]95%OpenAI Model Release Notes: Introducing two open-weight models: gpt-oss-120b and gpt-oss-20b (August 5, 2025)help.openai.com· Posted Sep 19, 2026· Starts Aug 5, 2025· 31% of score
OpenAI's[2] own model release notes date this entry August 5, 2025
- [2]90%gpt-oss-120b Model | OpenAI APIdevelopers.openai.com· Added Sep 22, 2026· 31% of score
OpenAI's[1] API model page for gpt-oss-120b lists configurable low/medium/high reasoning effort, full chain-of-thought, fine-tuning, function calling, web browsing and Python execution, text-only input and a 131,072 context window.
- [3]80%GPT-OSS - Wikipediaen.wikipedia.org· Added Sep 19, 2026· 31% of score
Says gpt-oss-120b and gpt-oss-20b open-weight reasoning models were released August 5, 2025 under Apache 2.0.[1]
- [4]85%Introducing gpt-oss | OpenAIweb.archive.org· Published Aug 5, 2025· 6% of score
Archived copy of OpenAI's[1][2] gpt-oss launch post: "gpt-oss-120b activates 5.1B parameters per token, while gpt-oss-20b activates 3.6B", weights "natively quantized in MXFP4", and benchmark comparisons with o4-mini.
Suggest a correction
Something missing or wrong? Say it in your own words: a link that backs this pin up, a different start or end date and why, or a fact it lacks or gets wrong. The AI checks it against this pin's sources, searches for better ones, and adds any page that backs you up. The pin's own sources still count most.
