Time to First Token

An LLM inference performance metric defined as the time interval between sending an API request and receiving the first output token. It reflects model preprocessing and scheduling overhead, and is a key measure of response speed from a user experience perspective.