Inference Speed

Warmed multilingual API generation. Lower real-time factor (RTF) is better.

Run Backend Calls Audio TTS time Median RTF p95 RTF Realtime speed
20261007-184048 native 312 1740.6s 2000.9s 1.134 1.272 0.87x
20261007-184048 nano 312 1730.4s 467.7s 0.271 0.315 3.70x
20261022-052037 nano 312 1730.6s 566.8s 0.326 0.443 3.05x
20261022-054748 nano 312 1730.6s 461.4s 0.271 0.329 3.75x
20261022-080929 nano 312 1730.6s 522.3s 0.309 0.373 3.31x
20261022-085849 nano 312 1730.6s 462.4s 0.272 0.328 3.74x
20261029-070100 nano 312 1730.6s 449.4s 0.262 0.317 3.85x

See DETAILS.md for per-language and per-mode results.