OpenAI's most powerful reasoning model. Pushes the frontier across coding, math, science, and visual perception. 20% fewer major errors than o1 on real-world tasks. First reasoning model with agentic tool use including web search and image understanding.