o3

o3 is a chat model developed by OpenAI as part of its o-series line of reasoning models. Unlike conventional chat completion models that respond immediately, o-series models allocate additional computation to an internal reasoning process before emitting a final answer, which shifts where latency appears in a request and changes how the model is billed and benchmarked. Independent measurements collected by Artificial Analysis put o3's output speed at roughly 138.8 tokens per second, with a time to first token of about 4,788 ms. That latency profile is characteristic of reasoning models: the initial wait covers deliberation before any visible output, after which generation proceeds at a normal streaming pace. Applications that display partial output to users should account for the multi-second gap before the first token arrives. Other specifications for o3 — including context window size and supported input modalities — are not tracked in our database, so consult OpenAI's model documentation or the individual provider's API reference for those details. In practice, o3 is used where accuracy on multi-step problems matters more than response latency: mathematical work, scientific analysis, debugging and code review, and planning steps inside agentic pipelines. Compared with OpenAI's non-reasoning chat models, it trades faster first-token response for additional deliberation, and compared with smaller o-series variants it sits at the higher-effort end of that trade-off. Providers and pricing for o3 vary; see the pricing table on this page for current options.