GPT-OSS
OpenAI's open-weight model line, its first open release since GPT-2, published in August 2025.
GPT-OSS is OpenAI’s family of open-weight models, released in August 2025 (the model card links an Aug 8 2025 paper) under Apache 2.0 in two sizes: gpt-oss-120b (117B parameters, 5.1B active) and gpt-oss-20b (21B parameters, 3.6B active), with the MoE weights quantized to MXFP4 so the 120B fits on one 80GB GPU and the 20B runs in 16GB. It is widely described as OpenAI’s first open-weight language model since GPT-2 and a response to the momentum of Llama, Qwen and DeepSeek; the model card does not say so, and OpenAI’s announcement could not be opened. GPT-OSS models are hosted on Hugging Face and widely used as bases for fine-tuning. See Open-weight vs. closed models.
Connections
GPT-OSS and everything one link away. Drag to move, scroll to zoom, click a node to open it.Open in the full graph →
What links here
Model family GPT
Company Hugging Face
Concept Open-weight vs. closed models
AI lab OpenAI