OpenAI o3 and o4-mini System Card

OpenAI o3 and o4-mini are models that combine state-of-the-art reasoning capabilities with the use of various tools, including web browsing, Python, image and file analysis, image generation, canvas, automation, file search, and memory. These models excel at solving complex math, coding, and scientific problems, and they extend their capabilities by directly using tools during their Chain of Thought, such as editing images or searching the web.

The OpenAI o-series models were trained using large-scale Reinforcement Learning on their chain of thought. This advanced reasoning capability provides a new avenue for improving the model's safety and robustness, and in particular supports Deliberative Alignment, which allows the model to reason about safety policies in context when responding to potentially unsafe prompts.

This release marks the first system card issued under Preparedness Framework Version 2. OpenAI's Safety Advisory Group (SAG) confirmed through its evaluation that o3 and o4-mini did not reach the 'High' threshold in any of the three tracked categories: biological and chemical capabilities, cybersecurity, and AI self-improvement.