Thinking Mode

Thinking Mode is the chain-of-thought pass before the visible answer. The official guide says it is on by default for V4, with default effort high. You can disable it or set reasoning_effort. The mapping I copied from the docs: requested medium or high becomes high; max stays max; low stays low.

Chat’s Expert Mode is the human toggle. On the API I send thinking enabled plus an effort. Temperature does not apply in thinking mode, per the same guide. I use high for refactors and max only when the agent is allowed to spend.

This is the grown-up form of DeepThink from DeepSeek-R1. Current hosts: Pro and Flash.