Saving reasoning tokens

  1. Are reasoning tokens preserved between tool calls within a single request?
  2. Are reasoning tokens preserved between requests in the same chat?

Up!

Cursor may preserve reasoning tokens across tool calls or between separate requests. In practice, this is largely determined by the underlying model/provider implementation as not all models support that, and may change over time.​

To help understand your needs better, when you say “saving reasoning tokens”, are you mostly thinking about cost/latency optimization, or about reusing prior reasoning across steps in a workflow?

This