Voice AI BenchmarksBy

MODEL DIRECTORY

Compare voice AI models

Independent benchmarks for 64 speech-to-text, text-to-speech and speech-to-speech models, re-measured every day.

Fastest final transcript (TTFS)#1 / 29
39ms
Qwen3 ASR 1.7bvia Baseten
Fastest first audio (TTFA)#1 / 32
60ms
vuivia Fluxions

64 models

Speech-to-Text: latency vs quality

Speed against accuracy for every measured STT model — the best sit toward the bottom left.

TTFS vs WEREach point is one measured STT model · 30-day averagesTime to Final Segment against Word Error Rate for every measured STT model.
Qwen3 ASR 1.7b — 39 ms, 2.9%STT RT v5 — 45 ms, 4.8%Qwen3 ASR Fast — 47 ms, 2.6%STT 1 — 66 ms, 7.1%Nova 3 — 85 ms, 5.3%Nova 2 — 87 ms, 6.7%Parakeet TDT 0.6B v3 — 97 ms, 11.1%Flux Multilingual — 98 ms, 7.5%Flux — 99 ms, 5.8%Scribe v2 Realtime — 122 ms, 4.4%Whisper Large v3 via Baseten — 131 ms, 4.5%Universal 3.6 Pro — 135 ms, 2.1%Ink 2 — 148 ms, 3.6%Universal 3.5 Pro — 168 ms, 2.4%Whisper Large v3 via Together AI — 200 ms, 7.4%Pulse — 205 ms, 4.6%Grok STT — 223 ms, 4.1%resonant-1 — 252 ms, 2.6%Default — 258 ms, 8.8%Linden 1 — 271 ms, 3.9%Gemini 3.5 Transcribe Live — 317 ms, 3.0%Daytona STT — 399 ms, 4.9%Voxtral Mini Transcribe Realtime 2602 — 411 ms, 5.4%GPT Realtime Whisper — 599 ms, 4.0%Solaria 1 — 638 ms, 6.1%GPT-4o mini Transcribe — 708 ms, 3.9%Chirp 3 — 768 ms, 3.2%GPT-4o Transcribe — 783 ms, 3.8%Chirp 2 — 908 ms, 3.9%

Text-to-Speech: latency vs quality

Speed against intelligibility for every measured TTS model — the best sit toward the bottom left.

TTFA vs WEREach point is one measured TTS model · 30-day averagesTime to First Audio against Word Error Rate for every measured TTS model.
vui — 60 ms, 2.1%Default — 63 ms, 1.9%Qwen3 TTS Fast — 68 ms, 1.8%TTS Flash 2 — 92 ms, 1.6%Qwen3 TTS 1.7b — 106 ms, 1.7%Simba 3.0 — 111 ms, 1.5%Palabra TTS v1 — 119 ms, 2.6%Simba 3.2 — 126 ms, 1.9%drama3-preview — 160 ms, 2.1%Flash v2.5 — 191 ms, 1.8%TTS 2 — 192 ms, 1.6%Eleven v4 Turbo — 201 ms, 1.7%Flux TTS — 205 ms, 1.7%TTS RT v1 — 254 ms, 2.4%TTS Rt v2 — 256 ms, 2.4%Mist v3 — 287 ms, 2.1%Sonic 3.5 — 293 ms, 1.5%Phantom Z 3.4 conversational — 296 ms, 1.8%Aura 2 — 297 ms, 1.8%Sonic 3.6 — 299 ms, 1.5%Falcon 2 — 304 ms, 1.7%Coda — 332 ms, 9.5%S2.1 Pro — 341 ms, 1.5%Grok TTS — 362 ms, 1.6%S1 — 389 ms, 1.8%Lightning v3.1 Pro — 392 ms, 1.5%Daytona TTS — 486 ms, 1.6%Aura 2 En — 573 ms, 2.0%Chirp 3 HD — 598 ms, 1.5%Qwen3 TTS Flash Realtime — 718 ms, 3.0%S2.1 Pro Free — 850 ms, 1.5%GPT-4o mini TTS — 2584 ms, 1.8%

About this directory

A placement always names its metric — #2 on TTFS is a different claim than #2 on WER — and is judged across every host serving the model. Models without enough qualifying samples in the last 30 days show no number at all.

Open-weight models can post different latency on different hosts because serving infrastructure and configuration differ, so each hosted endpoint is measured separately. The provider directory covers which organizations create models, host them, or do both.

Same datasets, prompts and metric definitions for every model, measured by Coval’s open-source runner. Full methodology on the overview.

Evaluate your own voice agent

Use Coval to test your production configuration, prompts and calls—not only the public benchmark endpoints.

Book a Demo