How long an LLM takes to begin responding after receiving a request — the metric that determines how responsive a model feels.
Time to first token (TTFT) is the delay between sending a prompt and receiving the first piece of the response. It covers network time, queueing, and "prefill" — the model processing your entire input before it can generate anything.
Users perceive responses under ~300ms as instant and over ~1s as sluggish, so TTFT is the key metric for chat interfaces, autocomplete and any interactive product. Long prompts increase TTFT because prefill scales with input length; reasoning models can add seconds of hidden thinking before the first visible token.
TTFT and throughput trade off differently by use case: interactive products should optimize TTFT first, batch pipelines can ignore it entirely.