GPT-OSS

OpenAI's first open-weight release since GPT-2, GPT-OSS brings frontier-style reasoning to models you can run on your own hardware. Both sizes are trained for chain-of-thought reasoning, function calling, and tool use, with adjustable reasoning effort so you can trade latency for depth.

The 20B model targets a single consumer GPU, while the 120B model is built for high-memory workstations and servers. Both ship in the efficient MXFP4 format and are released under a permissive license for commercial use.