Source: gpt-oss-120b model card (OpenAI, Hugging Face)
Model card for OpenAI's Apache 2.0 open-weight reasoning models gpt-oss-120b and gpt-oss-20b.
Contents
- What: Hugging Face model card
- Publisher: OpenAI
- Date: repository created 2025-08-04; accessed 2026-11-14
- Original: gpt-oss-120b
Summary
Two MoE reasoning models designed to fit on one 80GB GPU (120b) or 16GB of memory (20b).
Key claims
- gpt-oss-120b: 117B parameters, 5.1B active; gpt-oss-20b: 21B parameters, 3.6B active.
- Apache 2.0 licence.
- MoE weights post-trained in MXFP4, so 120b runs on a single H100/MI300X.
- Must be used with OpenAI’s harmony response format; outputs are wrong otherwise.
- Configurable reasoning effort (low/medium/high) and full chain-of-thought access (not meant for end users).
- Native function calling, browsing, Python execution and structured outputs.