Models & Architectures

Reasoning Model

An AI that "thinks" before answering — explicit step-by-step reasoning at inference time.

Also known as: o1,o3,thinking model

A reasoning model is an AI that explicitly thinks through problems before answering — generating an internal reasoning trace, possibly catching its own errors, then committing to a final response. OpenAI's o1/o3 series, GPT-5 Pro, Claude Opus 4.8 with "extended thinking", Gemini 3.5 Pro "Deep Think", and DeepSeek-R1 all use this approach. The trade-off: reasoning models are slower and pricier (thinking tokens are billed at output rates) but dramatically more accurate on hard problems. Claude Opus 4.7 without extended thinking scored 78% on GPQA Diamond; the same model with extended thinking scored 93%. Same weights, just more compute spent at inference time.

Read the full guide