Latency Budget Calculator
Response latency is the difference between 'wow' and 'weird'. Set each stage and see where your milliseconds go — and what to cut to get under 800ms.
280ms
120ms
320ms
160ms
90ms
Total response latency
970ms
0800ms target
Noticeable lag. Tighten the biggest stage.
Endpointing / VAD280ms
Speech-to-text120ms
LLM (time to first token)320ms
Text-to-speech160ms
Network / telephony90ms
Biggest lever right now: LLM (time to first token) at 320ms. Stream this stage and tune endpointing before adding features.