Voice AI BenchmarksBy

PROVIDERLAST 30 DAYS

official resource

NVIDIA voice AI models and benchmarks

NVIDIA lists 2 STT models in Coval. Its fastest dated STT result is Parakeet TDT 0.6B v3 on Together AI at mean 81 ms time to final segment (5th of 28) among STT systems, with 11.2% WER. Results cover the last 30 days. Last measured .

NVIDIA creates the open-weight Nemotron Speech Streaming and Parakeet TDT speech recognition models.

Measured models
2
STT
Best STT model#5 / 28
81ms
Parakeet TDT 0.6B v3

Overview

Nemotron Speech Streaming and Parakeet TDT expose model-card and weight identities while Together supplies the live serving path.

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
Nemotron 3.5 ASR Streaming, Parakeet TDT 0.6B v3

Speech-to-Text

Full STT dashboard

Ranked on Time to Final Segment against 28 measured models.

Time to Final Segmentms · lower is better · NVIDIA models markedEvery measured STT model on Time to Final Segment, with NVIDIA's models highlighted.
  1. #1Qwen3 ASR 1.7b39 ms
  2. #2Qwen3 ASR Fast46 ms
  3. #3STT RT v557 ms
  4. #4STT 165 ms
  5. #5Parakeet TDT 0.6B v381 ms
  6. #6Nova 389 ms
  7. #7Nova 292 ms
leaders plus NVIDIA models · 21 other models in the full table
Word Error Rate% · lower is better · NVIDIA models markedEvery measured STT model on Word Error Rate, with NVIDIA's models highlighted.
  1. #1Universal 3.5 Pro3.1%
  2. #2Qwen3 ASR Fast3.2%
  3. #3resonant-13.4%
  4. #4Gemini 3.5 Transcribe Live3.9%
  5. #5Chirp 34.1%
  6. #6Qwen3 ASR 1.7b4.1%
  7. #7Enhanced4.3%
  8. #29Parakeet TDT 0.6B v311.2%
  9. #30Nemotron 3.5 ASR Streaming15.6%
leaders plus NVIDIA models · 21 other models in the full table
Time to First Tokenms · lower is better · NVIDIA models markedEvery measured STT model on Time to First Token, with NVIDIA's models highlighted.
  1. #1Whisper Large v3via Baseten912 ms
  2. #2Qwen3 ASR 1.7b931 ms
  3. #3Universal 3.5 Pro1040 ms
  4. #4Flux1076 ms
  5. #5Flux Multilingual1163 ms
  6. #6Parakeet TDT 0.6B v31215 ms
  7. #7Whisper Large v3via Together AI1338 ms
  8. #15Nemotron 3.5 ASR Streaming1548 ms
leaders plus NVIDIA models · 18 other models in the full table
TTFS vs WER — NVIDIA vs every measured modelEach point is one measured STT model · NVIDIA models highlighted · 30-day averagesTime to Final Segment against Word Error Rate for every measured STT model, with NVIDIA's models highlighted.
Qwen3 ASR 1.7b — 39 ms, 4.1%Qwen3 ASR Fast — 46 ms, 3.2%STT RT v5 — 57 ms, 5.7%STT 1 — 65 ms, 4.4%Nova 3 — 89 ms, 6.2%Nova 2 — 92 ms, 7.9%Flux Multilingual — 98 ms, 8.8%Flux — 99 ms, 6.6%Ink 2 — 122 ms, 4.8%Whisper Large v3 via Baseten — 125 ms, 5.6%Scribe v2 Realtime — 133 ms, 5.5%Universal 3.5 Pro — 173 ms, 3.1%Grok STT — 207 ms, 4.6%Default via Speechmatics — 209 ms, 5.4%Pulse — 212 ms, 5.3%Default via Gradium — 246 ms, 10.0%resonant-1 — 264 ms, 3.4%Whisper Large v3 via Together AI — 296 ms, 8.5%Enhanced — 299 ms, 4.3%Gemini 3.5 Transcribe Live — 305 ms, 3.9%Voxtral Mini Transcribe Realtime 2602 — 400 ms, 6.2%GPT Realtime Whisper — 551 ms, 5.1%GPT-4o mini Transcribe — 690 ms, 4.9%GPT-4o Transcribe — 754 ms, 4.6%Chirp 3 — 776 ms, 4.1%Solaria 1 — 803 ms, 7.1%Chirp 2 — 873 ms, 4.9%Parakeet TDT 0.6B v3 — 81 ms, 11.2%
Benchmarked models with their Time to Final Segment over the last 30 days.
ModelHostTTFSRank
Nemotron 3.5 ASR StreamingTogether AI——
Parakeet TDT 0.6B v3Together AI81 ms5th of 28

How fast are NVIDIA's STT models?

Parakeet TDT 0.6B v3 on Together AI measures mean 81 ms time to final segment (5th of 28) among STT systems. Last measured 2026-09-15.

How accurate are NVIDIA's STT models?

Parakeet TDT 0.6B v3 on Together AI measures 11.2% word error rate (29th of 30) among STT systems. Last measured 2026-09-15. Nemotron 3.5 ASR Streaming on Together AI measures 15.6% word error rate (30th of 30) among STT systems. Last measured 2026-09-15.

Which NVIDIA model is fastest?

Its fastest dated STT result is Parakeet TDT 0.6B v3 on Together AI at mean 81 ms time to final segment (5th of 28) among STT systems, with 11.2% WER. Last measured 2026-09-15.

Limits of this comparison

Together hosts the endpoints measured by Coval.

  • Latency from Together should not be generalized to NVIDIA NIM, self-hosting or other optimized deployments of the same weights.

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