The GPT-OSS models were released on August 5, 2025. This release came after several delays, as OpenAI CEO Sam Altman had previously indicated in July that the company needed more time to “run additional safety tests and review high-risk areas,” following an earlier statement that the models would not be released in June.
The release represents a major milestone for OpenAI, as these are the first open-weight language models the company has released since GPT-2 in 2019. The timing appears strategic, positioning OpenAI to compete more directly with other companies like Meta, Mistral, and DeepSeek that have been releasing open-weight models in recent years. The delay in release was primarily attributed to safety considerations, with OpenAI conducting extensive safety training and testing, including adversarial fine-tuning evaluations to ensure the models meet their safety standards.
The models became available immediately upon release through platforms like Hugging Face and GitHub, with integration into Hugging Face Transformers as of version 4.55.0. Major cloud providers including Amazon, Microsoft Azure, and various GPU providers also made the models available for deployment on their platforms simultaneously with the release.
For more detailed information, see: GPT-oss vs o4-mini: Edge-Ready, On-Par Performance — Dependable, Not Mind-Blowing