The model also offers recent features for developers who incorporate OpenAI models of their software, including function calling, developer messages, and structured outputs. They can even select one in every of three reasoning effort options — low, medium, and high — to regulate power and latency to suit the use case. Nonetheless, unlike OpenAI o1, it doesn’t support vision capabilities. The corporate said that o3-mini is accessible within the Chat Completions API, Assistants API, and Batch API now, to pick out developers in API usage tiers 3-5.
Along with its performance, OpenAI touted the model’s safety. “Just like OpenAI o1, we discover that o3-mini significantly surpasses GPT-4o on difficult safety and jailbreak evaluations. Before deployment, we fastidiously assessed the security risks of o3-mini using the identical approach to preparedness, external red-teaming, and safety evaluations as o1.”
