Glossary
Latency is the time between a request and an agreed response point. A useful figure says what was measured and gives percentiles, not just an average.
Updated 21 Aug 2026
A latency figure without percentiles and a definition of what was timed is decoration. Time to first token and time to a complete, checked answer are very different numbers, and the flattering one is usually the one quoted.
Ask for a worked example on your own material, and the evidence needed to reproduce it.
We use “Latency” only where a product mechanism or an evaluation method backs it up, and we say when the behaviour depends on how a deployment is configured.
Pick a real piece of work, agree what a good answer looks like, then go through the results together.