Real numbers from real calls to the live API, measured 2026-10-08 10:51 UTC from a Darwin laptop, New York, home internet. Each figure is the median (half of calls were faster) with the 95th percentile (19 in 20 were faster).
| What | Median response time |
|---|---|
| Brand-new sentence (generated live) Text sent to POST /v1/speak that has never been spoken before; time until the audio link is ready. | 1.10 s (95% under 1.3 s) |
| WebSocket: time to first audio sentence Text sent over the live WebSocket; time until the first spoken sentence arrives. | 0.99 s (95% under 1.1 s) |
| Speech to text, short clip (about 4 s) POST /v1/listen with a four-second clip. | 0.72 s (95% under 1.0 s) |
| Speech to text, longer clip (about 15 s) POST /v1/listen with a fifteen-second clip. | 0.91 s (95% under 2.4 s) |
| Voice agent: first audio A customer question to an agent: time until the first spoken sentence of the answer is ready. | 1.10 s (95% under 1.9 s) |
| Voice agent: whole reply Question in, complete spoken answer ready. | 1.25 s (95% under 2.2 s) |
Re-run on request. Your own results depend on your location, text length and traffic.