The wait before an answer
Time to first token, or TTFT, describes the interval between sending a request and receiving the first visible text. AI Fast Now starts the clock at request dispatch. A connection event, metadata message or hidden reasoning event does not end that wait. This definition matters because a streaming API can be active before there is anything useful for a person to read.
Read the complete timeline
A useful record keeps dispatch, first event, first visible text, last visible text and completion separately. The interval before first text includes more than model computation: connection setup, network travel and service scheduling can all contribute. A single observation cannot separate those causes. We describe an observed wait, not internal server load.
Compare like with like
Compare the same model revision, prompt profile, output limit, reasoning configuration and probe location. A short health prompt and a long coding task answer different questions. A warm connection and a newly established connection can also produce different results. Those conditions belong in the experiment notes.
A practical reading rule
Start with sample count and freshness, then inspect the median and slower tail where sample size supports it. When no text arrives, retain a no-text or partial-stream outcome instead of assigning zero. At launch this website shows unavailable values until actual observations exist. A blank metric is more informative than a precise-looking number without evidence.