# gpt-oss — weights, license and what is actually open

Two open-weight text reasoning models with adjustable reasoning effort, structured output, and tool-use support.

Canonical: https://brightaifuture.com/open-models/gpt-oss
Format: model-family
Source publication: Not established
Bright publication: 2026-09-19
Substantive update: None recorded
Evidence and review: AI-assisted primary-source check on 2026-09-19; no independent model reproduction or license certification. OpenAI publishes extensive internal evaluations; independent reproductions are not incorporated in this record.

## Documented release

gpt-oss-120b and gpt-oss-20b

Organization: OpenAI

gpt-oss-120b and gpt-oss-20b release: 2025-08-05 (day precision). Source: https://openai.com/index/introducing-gpt-oss/

Parameters: 117B / 5.1B active; 21B / 3.6B active

Architecture: Transformer mixture-of-experts with alternating dense and locally banded sparse attention.

Modalities: text

Context: 128K tokens

## License and commercial use

Apache-2.0, with a separate usage policy

Apache-2.0 permits commercial use; deployers should also review the published usage policy.

## What is open

Weights: available. Both checkpoints are downloadable. Source: https://openai.com/index/introducing-gpt-oss/

Architecture: available. The model card publishes architecture and parameter details. Source: https://openai.com/index/introducing-gpt-oss/

Inference code: available. Reference PyTorch and Metal implementations and Harmony renderers are public. Source: https://openai.com/index/introducing-gpt-oss/

Training code: not-established. The complete training stack was not released. Source: https://openai.com/index/introducing-gpt-oss/

Training recipe: partial. Pre- and post-training approaches are described without a fully reproducible recipe. Source: https://openai.com/index/introducing-gpt-oss/

Data information: partial. OpenAI says the dataset is mostly English and text-only, focused on STEM, code, and general knowledge. Source: https://openai.com/index/introducing-gpt-oss/

Training data: not-established. No sufficiently specific public artifact was confirmed in this review. Source: https://openai.com/index/introducing-gpt-oss/

Evaluation: partial. A detailed model card and vendor evaluations are published. Source: https://openai.com/index/introducing-gpt-oss/

Commercial use: available. Apache-2.0 permits broad commercial use. Source: https://openai.com/index/introducing-gpt-oss/

## Uses and strengths described in sources

Tool use

Adjustable reasoning effort

Local deployment

Structured outputs

## Hardware and quantization

OpenAI states that 120b fits in 80 GB and 20b in 16 GB in the released MXFP4 format.

Official MXFP4

## Independent evidence

OpenAI publishes extensive internal evaluations; independent reproductions are not incorporated in this record.

## Limitations

Text-only

Training corpus is not released

Self-hosters own system-level safeguards

## Provenance and history

{}

## Original sources

- [Introducing gpt-oss](https://openai.com/index/introducing-gpt-oss/)
- [gpt-oss model card](https://cdn.openai.com/pdf/419b6906-9da6-406c-a19d-1bb078ac7637/oai_gpt-oss_model_card.pdf)

## Continue exploring

- [Open Models](https://brightaifuture.com/open-models)
- [Open intelligence thread](https://brightaifuture.com/threads/open)
- [Related evidence record](https://brightaifuture.com/discoveries/gpt-oss)
- [Running a reasoning model on your own machine](https://brightaifuture.com/discoveries/gpt-oss)