Time to first token
The elapsed time between a request and the first streamed model output received by the application or user.
Why it matters
It can shape perceived responsiveness, but it does not describe total completion time, task quality, or tool execution.
Measure it with
- Median
- Tail percentile
- Timeout rate
- Total task duration
Do not assume
A fast first token means the final result is useful, complete, or correct.