#1010th of 14 in Open models
gpt-oss
OpenAI's open-weight gpt-oss-120b and 20b models with reference inference code
Live demo ↗
(opens in a new tab)Documentation ↗
(opens in a new tab)Website ↗
(opens in a new tab)Repository on GitHub ↗
(opens in a new tab)
- Stars
- 20.5k
- License
- Apache-2.0
- Last commit
- Jul 2026
- Last release
- Jan 2026
- Language
- Python
Missed the latest GitHub check; these facts may be older.
OpenAI is hiring: 804 open roles
Overview
Repository for OpenAI's gpt-oss-120b (117B parameters, 5.1B active) and gpt-oss-20b (21B, 3.6B active) open-weight MoE models, with reference PyTorch, Triton and Metal inference code. It also includes browser and python tool implementations, a terminal chat client and an example Responses API server. Weights are downloaded from Hugging Face, and the models require the harmony response format.
Who it is for: Engineers evaluating or self-hosting OpenAI's open-weight reasoning models
Strengths
- Apache-2.0 license with weights on Hugging Face for both model sizes
- MXFP4 MoE weights: 120b fits one 80GB GPU, 20b runs within 16GB memory
- Configurable reasoning effort (low, medium, high) and full chain-of-thought output
- Runs via Transformers, vLLM, Ollama and LM Studio, plus native function calling
Weaknesses
- Torch, Triton and Metal implementations are reference code, not for production
- Models only work correctly with the harmony response format
- Torch reference needs at least 4x H100; Linux reference code requires CUDA
- Browser tool is educational and needs a You.com or Exa API key
What it needs
- RAM ≥ 16 GB
- GPU optional
- Needs Hugging Face Hub, openai-harmony, You.com or Exa API (browser tool)
- Models: gpt-oss-120b, gpt-oss-20b, MXFP4, SafeTensors
Also in Open models
See all 14| Rank | Project | Score |
|---|---|---|
| 1 | llmfitRust tool that checks which LLMs fit your RAM, CPU and GPU | 84 out of 100 |
| 2 | layaSingle-pass multilingual classifier for typed choice, score and yes/no decisions | 79 out of 100 |
| 3 | timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research | 72 out of 100 |
| #4 | unlimited-ocrOCR model that parses long documents and multi-page PDFs in one pass | 62 out of 100 |
| #5 | insightfaceFace detection, recognition and alignment toolbox with ArcFace and RetinaFace models | 57 out of 100 |