Watch AI reasoning models think — then keep the answer

Reasoning models — the o1 / R1 / Gemini-thinking generation — work differently from older chat models: they spend most of their output thinking before they write a single word of the answer. That hidden work is where the quality comes from, and normally you never see it. Grafy streams it live.

See the thinking, then the answer

Ask a hard question and the reasoning appears token by token in a Thinking… panel. The moment the model starts its real answer, the panel collapses to a tidy “Thought for 4s” you can re-open any time — transparency without the clutter.

Why watching it think matters

A visible chain of thought tells you why you got an answer: which assumption it made, which step it skipped. That is the difference between trusting a black box and checking its work.

It never gets cut off

Because a reasoning model can burn its whole budget thinking, a short token limit chops the reply mid-sentence. Grafy sizes the budget to finish the job, so the answer always lands — and you are only billed for the tokens actually used.

See a model think on your next question.