gpt-oss (OpenAI) overview and key facts (October 2026)

gpt-oss is OpenAI's family of open-weight language models for reasoning and agentic tasks, released under Apache 2.0 so companies can run them on their own hardware, including a smaller version for local use.

The family has two models: gpt-oss-120b with 117 billion parameters, 5.1 billion active, which fits on a single 80 GB GPU such as an H100, and gpt-oss-20b with 21 billion parameters, 3.6 billion active, which runs within 16 GB of memory. Both use OpenAI's harmony response format, support function calling, browsing and Python execution, expose their full chain of thought for debugging and can be fine-tuned. Together they had more than 11 million downloads on Hugging Face in the last month at our check.

• Runs reasoning on company hardwareThe 120b model fits on one data-center GPU, so companies can offer an internal reasoning assistant without sending data to a cloud API.
• Works offline on laptopsThe 20b model runs in 16 GB of memory. Field teams or sensitive departments can use a local assistant without network access.
• Adapts to specialist tasksFine-tuning lets developers train the model on internal examples. Domain vocabulary and formats improve for a specific process.

• Apache 2.0 license without copyleft terms, from a vendor that otherwise sells closed models.
• Requires the harmony response format; using it without that format does not work correctly.

OpenAI is based in San Francisco.

Self-hosting means your team runs, secures and updates the model, including GPU capacity, monitoring and patching. Check the license terms for commercial use and keep a record of which version you deploy. If outputs feed decisions about people, a human must stay in the loop. The raw chain of thought is not intended for end users.

No reviews yet. Be the first to share how your team uses it.

Every review shows its verification level. We read each one before it goes live. How we check reviews