Kokoro
Cheapest. Clean, plain read. Ignores cues.
Scores, speed, and the real rate.
The same trust rows as the open benchmark: a third-party quality score, our own measured latency, and the raw engine cost in the open.
- Quality Elo*
- 1060
- Measured latency
- 973ms measured 2026-06-10
- Languages
- 8 languages
- Follows [cues]
- Plain read, ignores cues
- Engine cost
- $0.62/M in · $0 out rate checked 2026-06-11
The provider’s published rate when we last checked. Rates move, and when they do we update this row. It’s here so you can weigh our flat pricing against the raw cost underneath, instead of taking our word for it.
- Rights
- Apache-2.0 model; commercial OK
* Quality Elo from the Artificial Analysis Speech Arena, retrieved June 10, 2026. It is a user-vote arena rating; the top model of all rated is Fun-Realtime-TTS at 1228.06.
Latency is our own wall-clock time to full audio, measured 2026-06-10 on the same path that serves the studio. A measurement, not a server SLA.
Every voice Kokoro ships.
Press play to hear each voice: real output from this engine, recorded unedited. What you hear is what it produced.
Warm American narrator
Steady American male
Soft British female
Low, gravelly male
Where Kokoro earns its place.
Drafts, high volume, cost-sensitive jobs.
Hear it in the studio.
Open the studio and generate real audio through Kokoro. No sign-up required to listen.