Time to first token (TTFT)
Time to first token (TTFT) is a performance metric that measures how long it takes from when a language model receives a prompt until it produces the first token of its response.
On this page
Source content checked . English translation published 2026-10-27.
In business automation
In a chat or voice agent, users mainly notice how long it takes for something to start happening. Time to first token (TTFT) is a performance metric that measures how long it takes from when a language model receives a prompt until it produces the first token of its response. It is part of the total latency, not the full time until the response is complete.
Examples
These examples illustrate a possible workflow. Adapt the connections and rules to your own systems.
Measuring response time in a service bot
A team building a service bot separately measures the time until the first word appears and the time until the response is complete. This shows whether the wait comes from the start of the response or from its length.
Relevant tools
- Anthropic
- This definition is from Anthropic's glossary.
What to check
- Confusing time to first token with total response time. A long response can start quickly and still take a while to finish.