Voice AI BenchmarksBy

PROVIDERLAST 30 DAYS

official resource

Baseten voice AI models and benchmarks

Baseten lists 3 STT and TTS models in Coval. Fastest dated mean latency over 30 days: STT: Qwen3 ASR 1.7b (dedicated inference) at 38 ms TTFS, with 2.9% WER. TTS: Qwen3 TTS 1.7b (dedicated inference) at 105 ms TTFA, with 1.9% WER. Last measured .

Baseten is an inference platform for deploying open-source and custom AI models on dedicated GPUs.

Measured models
3
STTTTS
Best STT model#1 / 29
38ms
Qwen3 ASR 1.7b
Best TTS model#5 / 31
105ms
Qwen3 TTS 1.7b

Overview

Baseten serves models created by others rather than training its own. Each measured endpoint is a streaming WebSocket deployment from its model library, with hardware and concurrency chosen per deployment.

Every model below is measured daily on the same fixed inputs and ranked against the full field, never blended into a company score.

Model lineup

Speech-to-Text
serves Whisper Large v3, Qwen3 ASR 1.7b
Text-to-Speech
serves Qwen3 TTS 1.7b

Speech-to-Text

Full STT dashboard

Ranked on Time to Final Segment against 29 measured models.

Time to Final Segmentms · lower is better · Baseten models markedEvery measured STT model on Time to Final Segment, with Baseten's models highlighted.
  1. #1Qwen3 ASR 1.7b38 ms
  2. #2STT RT v543 ms
  3. #3Qwen3 ASR Fast49 ms
  4. #4STT 169 ms
  5. #5Nova 385 ms
  6. #6Nova 286 ms
  7. #7Parakeet TDT 0.6B v397 ms
  8. #12Whisper Large v3via Baseten129 ms
leaders plus Baseten models · 21 other models in the full table
Word Error Rate% · lower is better · Baseten models markedEvery measured STT model on Word Error Rate, with Baseten's models highlighted.
  1. #1Universal 3.6 Pro2.2%
  2. #2Universal 3.5 Pro2.4%
  3. #3Qwen3 ASR Fast2.6%
  4. #4resonant-12.7%
  5. #5Qwen3 ASR 1.7b2.9%
  6. #6Gemini 3.5 Transcribe Live3.0%
  7. #7Chirp 33.2%
  8. #17Whisper Large v3via Baseten4.5%
leaders plus Baseten models · 23 other models in the full table
Time to First Tokenms · lower is better · Baseten models markedEvery measured STT model on Time to First Token, with Baseten's models highlighted.
  1. #1Whisper Large v3via Baseten922 ms
  2. #2Qwen3 ASR 1.7b945 ms
  3. #3Universal 3.5 Pro1020 ms
  4. #4STT 11063 ms
  5. #5Universal 3.6 Pro1078 ms
  6. #6Flux1094 ms
  7. #7Flux Multilingual1170 ms
leaders plus Baseten models · 20 other models in the full table
TTFS vs WER — Baseten vs every measured modelEach point is one measured STT model · Baseten models highlighted · 30-day averagesTime to Final Segment against Word Error Rate for every measured STT model, with Baseten's models highlighted.
STT RT v5 — 43 ms, 4.8%Qwen3 ASR Fast — 49 ms, 2.6%STT 1 — 69 ms, 4.2%Nova 3 — 85 ms, 5.3%Nova 2 — 86 ms, 6.6%Parakeet TDT 0.6B v3 — 97 ms, 11.1%Flux Multilingual — 98 ms, 7.5%Flux — 99 ms, 5.8%Scribe v2 Realtime — 118 ms, 4.3%Universal 3.6 Pro — 125 ms, 2.2%Universal 3.5 Pro — 145 ms, 2.4%Ink 2 — 152 ms, 3.5%Whisper Large v3 via Together AI — 182 ms, 7.3%Grok STT — 191 ms, 4.2%Pulse — 204 ms, 4.6%resonant-1 — 246 ms, 2.7%Default — 262 ms, 8.6%Linden 1 — 274 ms, 3.9%Gemini 3.5 Transcribe Live — 343 ms, 3.0%Daytona STT — 398 ms, 4.9%Solaria 1 — 430 ms, 6.2%Voxtral Mini Transcribe Realtime 2602 — 436 ms, 5.4%GPT Realtime Whisper — 557 ms, 4.0%GPT-4o mini Transcribe — 721 ms, 3.9%Chirp 3 — 766 ms, 3.2%GPT-4o Transcribe — 781 ms, 3.8%Chirp 2 — 898 ms, 3.9%Qwen3 ASR 1.7b — 38 ms, 2.9%Whisper Large v3 via Baseten — 129 ms, 4.5%
Benchmarked models with their Time to Final Segment over the last 30 days.
ModelHostTTFSRank
Whisper Large v3Baseten129 ms12th of 29
Qwen3 ASR 1.7bBaseten38 ms1st of 29

Text-to-Speech

Full TTS dashboard

Ranked on Time to First Audio against 31 measured models.

Time to First Audioms · lower is better · Baseten models markedEvery measured TTS model on Time to First Audio, with Baseten's models highlighted.
  1. #1vui44 ms
  2. #2Default63 ms
  3. #3Qwen3 TTS Fast68 ms
  4. #4TTS Flash 281 ms
  5. #5Qwen3 TTS 1.7b105 ms
  6. #6Simba 3.0110 ms
  7. #7Palabra TTS120 ms
leaders plus Baseten models · 24 other models in the full table
Word Error Rate% · lower is better · Baseten models markedEvery measured TTS model on Word Error Rate, with Baseten's models highlighted.
  1. #1Sonic 3.61.5%
  2. #2Sonic 3.51.6%
  3. #3Lightning v3.1 Pro1.6%
  4. #4S2.1 Pro Free1.6%
  5. #5Simba 3.01.6%
  6. #6S2.1 Pro1.6%
  7. #7Chirp 3 HD1.6%
  8. #18Qwen3 TTS 1.7b1.9%
leaders plus Baseten models · 23 other models in the full table
TTFA vs WER — Baseten vs every measured modelEach point is one measured TTS model · Baseten models highlighted · 30-day averagesTime to First Audio against Word Error Rate for every measured TTS model, with Baseten's models highlighted.
vui — 44 ms, 2.1%Default — 63 ms, 1.9%Qwen3 TTS Fast — 68 ms, 1.9%TTS Flash 2 — 81 ms, 1.7%Simba 3.0 — 110 ms, 1.6%Palabra TTS — 120 ms, 2.7%Simba 3.2 — 124 ms, 2.0%drama3-preview — 168 ms, 1.7%TTS 2 — 177 ms, 1.7%Eleven v4 Turbo — 195 ms, 1.7%Flash v2.5 — 195 ms, 1.9%Flux TTS — 204 ms, 1.8%Phantom Z 3.4 conversational — 209 ms, 2.3%Falcon 2 — 237 ms, 1.7%TTS RT v1 — 256 ms, 2.6%TTS Rt v2 — 259 ms, 2.5%Airy TTS — 263 ms, 1.7%Mist v3 — 286 ms, 2.1%Sonic 3.5 — 289 ms, 1.6%Sonic 3.6 — 293 ms, 1.5%Aura 2 — 298 ms, 1.9%Coda — 330 ms, 9.6%S2.1 Pro — 351 ms, 1.6%Grok TTS — 351 ms, 1.7%Lightning v3.1 Pro — 399 ms, 1.6%Daytona TTS — 469 ms, 1.8%Aura 2 En — 558 ms, 2.1%Chirp 3 HD — 590 ms, 1.6%Qwen3 TTS Flash Realtime — 722 ms, 3.1%S2.1 Pro Free — 829 ms, 1.6%Qwen3 TTS 1.7b — 105 ms, 1.9%
Benchmarked models with their Time to First Audio over the last 30 days.
ModelHostTTFARank
Qwen3 TTS 1.7bBaseten105 ms5th of 31

How fast are Baseten's STT and TTS models?

Qwen3 ASR 1.7b (dedicated inference) measures mean 38 ms time to final segment (1st of 29) among STT systems. Last measured 2026-10-24. Whisper Large v3 (dedicated inference) measures mean 129 ms time to final segment (12th of 29) among STT systems. Last measured 2026-10-24. Qwen3 TTS 1.7b (dedicated inference) measures mean 105 ms time to first audio (5th of 31) among TTS systems. Last measured 2026-10-24.

How accurate are Baseten's STT and TTS models?

Qwen3 ASR 1.7b (dedicated inference) measures 2.9% word error rate (5th of 31) among STT systems. Last measured 2026-10-24. Whisper Large v3 (dedicated inference) measures 4.5% word error rate (17th of 31) among STT systems. Last measured 2026-10-24. Qwen3 TTS 1.7b (dedicated inference) measures 1.9% word error rate (18th of 31) among TTS systems. Last measured 2026-10-24.

Which Baseten model is fastest?

Its fastest dated STT result is Qwen3 ASR 1.7b (dedicated inference) at mean 38 ms time to final segment (1st of 29) among STT systems, with 2.9% WER. Last measured 2026-10-24. Its fastest dated TTS result is Qwen3 TTS 1.7b (dedicated inference) at mean 105 ms time to first audio (5th of 31) among TTS systems, with 1.9% WER. Last measured 2026-10-24.

Limits of this comparison

Coval measures three open-weight models served from Baseten dedicated endpoints: Whisper Large v3 and Qwen3-ASR 1.7B for transcription, and Qwen3-TTS 1.7B for synthesis.

  • Dedicated endpoints run on hardware reserved for a single customer, so their latency is not ranked against shared APIs here; accuracy is compared across the full field. Results describe Coval's deployments, not every configuration Baseten's library offers.

Official resources

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