gpt-oss-120b & gpt-oss-20b Model Card

Source: OpenAI Date: 2025-08-05 URL: https://openai.com/index/gpt-oss-model-card

Summary

Summary

The model card for gpt-oss-120b and gpt-oss-20b documented the capabilities, safety evaluations, intended use cases, and known limitations of OpenAI’s first open-weight models released in August 2025. The card covered both the flagship 120B parameter model and the smaller 20B model, describing their performance characteristics, safety properties, and the conditions under which OpenAI was releasing the weights.

Implications

Open-weight/model line thread. The gpt-oss model card is significant documentation for OpenAI’s open-weight strategy: it establishes what OpenAI considers acceptable use of open weights, what safety properties were evaluated before release, and what the models are not intended for. The dual-size release (120B and 20B) reflects lessons from Meta’s tiered Llama releases — different parameter counts serve different use cases and hardware budgets. The safety evaluation section in the model card is particularly important for open-weight models, because once weights are released, OpenAI has no enforcement mechanism — so the card’s safety analysis is both an evaluation record and a statement of what OpenAI considered acceptable risk for unrestricted distribution. The release marks a strategic inflection in OpenAI’s relationship to the open-weight ecosystem.