← back to the directory
Model Inference & Serving

gpt-oss

1OpenAI 2open 3weight 4reasoning 5model 6release

the six Ws · specification

W1 Who

Released by OpenAI as its first open weight model family since GPT-2.

W2 What

gpt-oss is a pair of open weight reasoning models, 120b and 20b, built for strong reasoning and agentic performance.

W3 Where

Weights and code are hosted on GitHub at openai/gpt-oss and on the Hugging Face Hub.

W4 When

Released in August 2025.

W5 Why

It gives developers a locally runnable, permissively licensed alternative to OpenAI's closed API models for reasoning heavy workloads.

W6 With

Uses a mixture of experts architecture and runs via Transformers, vLLM, Ollama or llama.cpp.

W7 Watch

Apache 2.0 licensed, open weights, released and maintained by OpenAI, roughly 20k GitHub stars.

open-weightsreasoning-modelapache-licensedopenaimixture-of-experts

for agents & scripts

Reading this as a machine? Query it directly.

Search is open JSON - no key. Report telemetry after using a tool and it feeds that tool’s Proof Score. Or speak MCP to /mcp and discover tools mid-loop.