# openai/gpt-oss

gpt-oss-120b and gpt-oss-20b are two open-weight language models by OpenAI

Repository: https://github.com/openai/gpt-oss
Canonical: https://ross.abutalabs.com/products/gpt-oss
Homepage: https://openai.com/open-models
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-07-24T18:56:01+00:00

## Health v2 (maintenance only)
Score: 71/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 94, release rhythm 65, longevity 31
- inputs: {"age_days": 436, "days_push": 40, "days_rel": 232, "gap_med": 11, "n_releases_24m": 6}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 20347, forks 2140 (observed 2026-08-28T04:11:29.993883+00:00)

## What it is
OpenAI's gpt-oss repository hosts reference implementations and tooling for two open-weight language models, gpt-oss-120b and gpt-oss-20b, released under Apache 2.0. The models are Mixture-of-Experts reasoning models with configurable reasoning effort, full chain-of-thought access, agentic tool use, and MXFP4 quantization enabling single-GPU or local deployment.

## Use cases
- run an open-weight LLM locally on a laptop or single GPU
- self-host a reasoning model for agentic tasks with function calling
- fine-tune an open model for a custom domain
- deploy a 120B language model on one 80GB H100
- get full chain-of-thought output for debugging model reasoning
- run a small 20B model within 16GB of memory
- build a chatbot with an Apache-2.0 licensed model

## When to choose
- you need an open-weight, commercially usable LLM without copyleft restrictions
- you want configurable reasoning effort and visible chain-of-thought
- you need agentic capabilities like tool calling and code execution in a self-hosted model
- you want to fine-tune a strong reasoning model on your own data

## When to avoid
- you need multimodal image or audio input from the model
- you lack a GPU or sufficient memory for even the 20B variant
- you want a plug-and-play API without managing inference infrastructure
- your workflow cannot adopt the required harmony response format

## Facets
- artifact type: library
- maturity: stable
- function: llm-inference, machine-learning, agent-framework, sdk
- domain: large-language-models, artificial-intelligence, deep-learning
- platform: python, cross-platform, self-hosted
- tags: open-weight-models, moe, mxfp4-quantization, reasoning-models, harmony-format, fine-tuning, hugging-face, local-inference, ai-agents, gpu

## Member repositories
- openai/gpt-oss (main) score 71

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:11:29.993883+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T16:59:01.977957+00:00, confidence not recorded.
  - readme: https://github.com/openai/gpt-oss (fetched 2026-08-28T04:11:29.993883+00:00, sha 5525d2ee6ffa)
  - homepage: https://openai.com/open-models (fetched 2026-08-29T07:57:40.368370+00:00, sha 735447dca92a)
  - site_page: https://openai.com/about (fetched 2026-08-29T07:57:40.377577+00:00, sha de4b627b33e3)
  - site_page: https://developers.openai.com/api/docs (fetched 2026-08-29T07:57:40.384553+00:00, sha d617c5215f2a)
  - site_page: https://cookbook.openai.com/articles/openai-harmony (fetched 2026-08-29T07:57:40.379348+00:00, sha 85f49884d584)
  - site_page: https://cookbook.openai.com/topic/gpt-oss (fetched 2026-08-29T07:57:40.382228+00:00, sha beeaf06e9813)
  - site_page: https://developers.openai.com/ (fetched 2026-08-29T07:57:40.387594+00:00, sha 7d240c906018)
- Data as of 2026-08-30T08:39:29.467469+00:00.