OpenAI has released an official guide detailing function calling and tool use for its o3 and o4-mini models. These models incorporate tool use within their chain of thought, enhancing reasoning capabilities and function calling performance without the need for explicit chain-of-thought prompts. The guide provides instructions on crafting prompts and schemas that reduce errors, ensure accurate function calls, and maintain coherent reasoning across multiple calls. Observers have noted improvements in the o3 model's reasoning process, including multiple distinct thinking phases and the ability to self-correct repetitive search loops. The guide is available on GitHub and aims to assist developers in optimizing the performance of these AI models.