Learn
What is time to first token (TTFT)?
Last reviewed July 16, 2026
Time to first token (TTFT) is the delay between sending a request and the first token of the model's reply arriving. It covers network transit, queueing, and prompt processing: the model must read your whole input before writing anything. Once streaming starts, throughput in tokens per second takes over; TTFT is the silence before, and it dominates how fast a model feels in interactive use.
What drives TTFT
Prompt length is the big one: processing input scales with its size, so a 100K-token context can add seconds before output starts. Provider load and queueing add variance. Reasoning models add think-time by design: they can spend seconds to minutes reasoning before the first visible token, which is a capability trade, not an infrastructure fault.
TTFT vs throughput
Two different numbers: TTFT measures responsiveness, tokens per second measures generation rate. Voice interfaces and chat live and die on TTFT; long document generation cares about throughput. Comparing models on one number misleads: a model can start instantly and write slowly, or think for ten seconds and then write very fast.
idapt measures TTFT, throughput, and uptime for every model on production traffic and publishes the live numbers on the model status board, with a live ranking of the fastest models.
Frequently asked
See it in practice
The fastest way to understand it is to use it. Start free, no credit card.
Try idapt free