o3

By OpenAI

Model family: GPT

o3 is built to plan, verify intermediate steps, and produce structured answers that downstream systems can trust. It keeps long conversations coherent, reads multi document prompts, and handles repository level code tasks with clear explanations. The API supports function calls, streaming, and schema true JSON, so agents can search, run tools, or retrieve data in the loop while keeping outputs parseable. You can trade speed for quality by allowing a larger internal thinking budget when the problem is hard. Teams often pair o3 with a faster sibling for routine prompts and reserve o3 for analysis, policy aware drafting, analytics, and complex software work where correctness matters. Alignment targets safe, straightforward language that avoids over refusal while minimizing hallucinations, which makes o3 a dependable core for production workflows.

New Text Gen 7

Released: October 15, 2025