Introducing gpt-oss

Details

OpenAI has released gpt-oss, a family of open-weight models in two sizes — gpt-oss-20b and gpt-oss-120b — available under an open license. This is a significant strategic shift: OpenAI, long the strongest proponent of closed-source AI development, is now competing directly in the open-weights space dominated by Meta's Llama series, Mistral, and others. The 120B parameter model in particular positions gpt-oss as a heavyweight open-weight contender. Developers and enterprises can now run OpenAI-architecture models locally or in private cloud environments, which directly addresses the data privacy and regulatory concerns that have limited ChatGPT API adoption in some sectors. From a competitive standpoint, this puts pressure on Meta's Llama 4 and Mistral's commercial offerings. It also signals that OpenAI believes it can maintain commercial advantage through its hosted products (ChatGPT, API) even while releasing capable open weights — a bet that alignment and product experience matter more than model access. The gpt-oss release comes with an official model card, suggesting OpenAI is taking responsibility for downstream use of these weights, which will be important for regulatory discussions around open-weight model liability.