Voice AI BenchmarksBy

ALIBABASPEECH-TO-TEXT68,345 SAMPLES / 30 DAYSLAST RUN OCT 16, 2026, 00:00 UTC

official resource

Qwen3 ASR 1.7b speech-to-text benchmarks

Qwen3 ASR 1.7b, hosted by Baseten (dedicated inference), measures mean 39 ms time to final segment (1st of 29) and 2.9% word error rate (5th of 31) among STT systems. Results cover the last 30 days. Coval benchmarks hosted endpoints daily using a consistent evaluation methodology. Last measured .

Qwen3-ASR 1.7B is Alibaba's open-weight speech recognition model in the Qwen3 family, released under the Apache 2.0 license.

Time to Final Segment#1 / 29
39ms
Word Error Rate#5 / 31
2.9%
Time to First Token#2 / 27
946ms

Overview

The 1.7B-parameter model transcribes 30+ languages and 20+ Chinese dialects with automatic language detection; Baseten's streaming harness adds a configurable partial-transcript cadence and voice-activity detection.

Qwen3 ASR 1.7b is tested every day on fixed public audio — clean, accented, noisy, reverberant, far-field, clipped and phone-codec speech — for transcription accuracy and streaming latency.

Technical specifications

Made by
Alibaba
Hosted by
Baseten
Source
Dedicated inference
Licensing
Open-weight
Deployment
On-prem
Region
US
Features
Multilingual, VAD

How Qwen3 ASR 1.7b ranks

Full STT dashboard
Time to Final Segment — every measured STT modelMilliseconds · lower is better · 30-day averageEvery STT model ranked on Time to Final Segment, with Qwen3 ASR 1.7b highlighted.
  1. #1Qwen3 ASR 1.7b39 ms
  2. #2STT RT v545 ms
  3. #3Qwen3 ASR Fast47 ms
  4. #4STT 168 ms
  5. #5Nova 385 ms
  6. #6Nova 287 ms
  7. #7Parakeet TDT 0.6B v396 ms
  8. #8Flux Multilingual98 ms
  9. #9Flux99 ms
  10. #10Scribe v2 Realtime120 ms
  11. #11Whisper Large v3via Baseten132 ms
  12. #12Universal 3.6 Pro134 ms
Show all 29 models
  1. #13Ink 2149 ms
  2. #14Universal 3.5 Pro165 ms
  3. #15Whisper Large v3via Together AI192 ms
  4. #16Pulse206 ms
  5. #17Grok STT222 ms
  6. #18resonant-1251 ms
  7. #19Default262 ms
  8. #20Linden 1275 ms
  9. #21Gemini 3.5 Transcribe Live323 ms
  10. #22Daytona STT399 ms
  11. #23Voxtral Mini Transcribe Realtime 2602431 ms
  12. #24Solaria 1534 ms
  13. #25GPT Realtime Whisper598 ms
  14. #26GPT-4o mini Transcribe712 ms
  15. #27Chirp 3769 ms
  16. #28GPT-4o Transcribe785 ms
  17. #29Chirp 2910 ms
median of all models · 192 ms
STT models over the last 30 days, ranked on Time to Final Segment.
#ModelHostTTFSWERTTFTSamples
1Qwen3 ASR 1.7bBaseten39 ms946 ms13,669
2STT RT v5Soniox45 ms1519 ms25,378
3Qwen3 ASR FastNari Labs47 ms1752 ms26,672
4STT 1Inworld AI68 ms1167 ms24,769
5Nova 3Deepgram85 ms1436 ms26,023
6Nova 2Deepgram87 ms1418 ms25,978
7Parakeet TDT 0.6B v3Together AI96 ms1222 ms26,478
8Flux MultilingualDeepgram98 ms1166 ms26,026
9FluxDeepgram99 ms1091 ms26,027
10Scribe v2 RealtimeElevenLabs120 ms2154 ms26,660
11Whisper Large v3Baseten132 ms925 ms13,667
12Universal 3.6 ProAssemblyAI134 ms1074 ms26,636
13Ink 2Cartesia149 ms1985 ms26,673
14Universal 3.5 ProAssemblyAI165 ms1023 ms26,674
15Whisper Large v3Together AI192 ms1246 ms26,487
16PulseSmallest206 ms1977 ms26,674
17Grok STTSpaceXAI222 ms—26,596
18resonant-1Reson8251 ms—26,673
19DefaultGradium262 ms1991 ms26,671
20Linden 1Speechmatics275 ms1539 ms26,615
21Gemini 3.5 Transcribe LiveGemini323 ms1574 ms26,580
22Daytona STTGuava399 ms5690 ms6,088
23Voxtral Mini Transcribe Realtime 2602Mistral431 ms1858 ms25,852
24Solaria 1Gladia534 ms1678 ms26,491
25GPT Realtime WhisperOpenAI598 ms1825 ms26,641
26GPT-4o mini TranscribeOpenAI712 ms—26,641
27Chirp 3Google769 ms5958 ms26,659
28GPT-4o TranscribeOpenAI785 ms—26,643
29Chirp 2Google910 ms6101 ms26,670
—Nemotron 3.5 ASR StreamingTogether AI—1547 ms26,536
—Universal StreamingAssemblyAI—1510 ms26,644
Under-sampled models are excluded; tied models share a place. Dotted WER values split into substitutions, deletions and insertions on hover or tap.

Qwen3 ASR 1.7b runs on dedicated inference, ranked here against shared endpoints.

Latency vs accuracy

TTFS vs WEREach point is one measured model · Qwen3 ASR 1.7b highlighted · 30-day averagesTime to Final Segment against Word Error Rate for every measured STT model, with Qwen3 ASR 1.7b highlighted.
STT RT v5 — 45 ms, 4.8%Qwen3 ASR Fast — 47 ms, 2.6%STT 1 — 68 ms, 6.8%Nova 3 — 85 ms, 5.3%Nova 2 — 87 ms, 6.7%Parakeet TDT 0.6B v3 — 96 ms, 11.1%Flux Multilingual — 98 ms, 7.5%Flux — 99 ms, 5.8%Scribe v2 Realtime — 120 ms, 4.3%Whisper Large v3 via Baseten — 132 ms, 4.4%Universal 3.6 Pro — 134 ms, 2.2%Ink 2 — 149 ms, 3.6%Universal 3.5 Pro — 165 ms, 2.4%Whisper Large v3 via Together AI — 192 ms, 7.4%Pulse — 206 ms, 4.6%Grok STT — 222 ms, 4.2%resonant-1 — 251 ms, 2.6%Default — 262 ms, 8.7%Linden 1 — 275 ms, 3.9%Gemini 3.5 Transcribe Live — 323 ms, 3.0%Daytona STT — 399 ms, 4.9%Voxtral Mini Transcribe Realtime 2602 — 431 ms, 5.4%Solaria 1 — 534 ms, 6.1%GPT Realtime Whisper — 598 ms, 4.1%GPT-4o mini Transcribe — 712 ms, 3.9%Chirp 3 — 769 ms, 3.2%GPT-4o Transcribe — 785 ms, 3.7%Chirp 2 — 910 ms, 3.9%Qwen3 ASR 1.7b — 39 ms, 2.9%

Where the errors come from

WER compositionQwen3 ASR 1.7b's Word Error Rate split by error type · 30-day averageQwen3 ASR 1.7b's WER split into substitutions, deletions and insertions.
  • Qwen3 ASR 1.7b2.9%
SubstitutionsDeletionsInsertions

Averages and tail latency

Averages hide slow outliers — these are the distributions behind each figure.

Qwen3 ASR 1.7b latency distributionMilliseconds · shared axis across metrics · last 30 daysQwen3 ASR 1.7b's latency percentiles per metric: p25–p75 band, p50 tick, whisker to max.
  • Time to Final Segmentp50 23 ms · Max 369 ms
  • Time to First Tokenp50 1054 ms · Max 4883 ms
band p25–p75 · tick p50 · whisker to max with p90 and p95 stops
Average and percentile values per metric, with the number of samples behind each row.
MetricAveragep25p50p75p90p95MaxSamples
Time to Final Segment39 ms21 ms23 ms33 ms98 ms115 ms369 ms13,669
Word Error Rate2.9%0.0%0.0%4.8%11.1%15.4%100.0%13,669
Time to First Token946 ms655 ms1054 ms1074 ms1457 ms1757 ms4883 ms13,669

Last 30 days

Daily medians from the same measurement runs · gaps are days without qualifying runs.

Time to Final Segment — daily p50Line p50 · band p25–p75 · UTC daysQwen3 ASR 1.7b's daily median Time to Final Segment over the last 30 days.
Qwen3 ASR 1.7b · Sep 17: 21 msQwen3 ASR 1.7b · Sep 18: 22 msQwen3 ASR 1.7b · Sep 19: 22 msQwen3 ASR 1.7b · Sep 20: 22 msQwen3 ASR 1.7b · Sep 21: 21 msQwen3 ASR 1.7b · Sep 22: 21 msQwen3 ASR 1.7b · Sep 23: 33 msQwen3 ASR 1.7b · Sep 24: 24 msQwen3 ASR 1.7b · Sep 25: 23 msQwen3 ASR 1.7b · Sep 26: 23 msQwen3 ASR 1.7b · Sep 27: 32 msQwen3 ASR 1.7b · Sep 28: 21 msQwen3 ASR 1.7b · Sep 29: 33 msQwen3 ASR 1.7b · Sep 30: 31 msQwen3 ASR 1.7b · Oct 1: 32 msQwen3 ASR 1.7b · Oct 2: 31 msQwen3 ASR 1.7b · Oct 3: 21 msQwen3 ASR 1.7b · Oct 4: 24 msQwen3 ASR 1.7b · Oct 5: 21 msQwen3 ASR 1.7b · Oct 6: 21 msQwen3 ASR 1.7b · Oct 7: 20 msQwen3 ASR 1.7b · Oct 8: 24 msQwen3 ASR 1.7b · Oct 9: 21 msQwen3 ASR 1.7b · Oct 10: 21 msQwen3 ASR 1.7b · Oct 11: 21 msQwen3 ASR 1.7b · Oct 12: 20 msQwen3 ASR 1.7b · Oct 13: 21 msQwen3 ASR 1.7b · Oct 14: 25 msQwen3 ASR 1.7b · Oct 15: 26 msQwen3 ASR 1.7b · Oct 16: 20 ms
Word Error Rate — daily averageDaily average · UTC daysQwen3 ASR 1.7b's daily Word Error Rate over the last 30 days.
Qwen3 ASR 1.7b · Sep 17: 3.3%Qwen3 ASR 1.7b · Sep 18: 3.8%Qwen3 ASR 1.7b · Sep 19: 3.2%Qwen3 ASR 1.7b · Sep 20: 3.0%Qwen3 ASR 1.7b · Sep 21: 3.4%Qwen3 ASR 1.7b · Sep 22: 3.0%Qwen3 ASR 1.7b · Sep 23: 2.9%Qwen3 ASR 1.7b · Sep 24: 3.7%Qwen3 ASR 1.7b · Sep 25: 3.1%Qwen3 ASR 1.7b · Sep 26: 3.1%Qwen3 ASR 1.7b · Sep 27: 2.7%Qwen3 ASR 1.7b · Sep 28: 3.1%Qwen3 ASR 1.7b · Sep 29: 2.9%Qwen3 ASR 1.7b · Sep 30: 3.1%Qwen3 ASR 1.7b · Oct 1: 2.7%Qwen3 ASR 1.7b · Oct 2: 3.5%Qwen3 ASR 1.7b · Oct 3: 3.3%Qwen3 ASR 1.7b · Oct 4: 3.4%Qwen3 ASR 1.7b · Oct 5: 3.6%Qwen3 ASR 1.7b · Oct 6: 3.1%Qwen3 ASR 1.7b · Oct 7: 3.5%Qwen3 ASR 1.7b · Oct 8: 3.5%Qwen3 ASR 1.7b · Oct 9: 3.3%Qwen3 ASR 1.7b · Oct 10: 3.2%Qwen3 ASR 1.7b · Oct 11: 3.3%Qwen3 ASR 1.7b · Oct 12: 3.4%Qwen3 ASR 1.7b · Oct 13: 3.6%Qwen3 ASR 1.7b · Oct 14: 3.4%Qwen3 ASR 1.7b · Oct 15: 3.0%Qwen3 ASR 1.7b · Oct 16: 3.5%

Time to Final Segment by dataset

Qwen3 ASR 1.7b by test conditionMilliseconds · lower is better · best condition firstQwen3 ASR 1.7b's Time to Final Segment on each benchmark dataset.
  1. WildASR accents24 ms
  2. WildASR phone codec27 ms
  3. PipeCat (production)32 ms
  4. WildASR clean33 ms
  5. WildASR noise gaps39 ms
  6. WildASR far-field83 ms
  7. WildASR clipping105 ms
Time to Final Segment per dataset, with the number of samples behind each figure.
DatasetTTFSSamples
ProductionPipeCat32 ms7,195
AccentsWildASR24 ms720
CleanWildASR33 ms2,874
ClippingWildASR105 ms720
Far-fieldWildASR83 ms720
Noise gapsWildASR39 ms720
Phone codecWildASR27 ms720

Strongest condition: WildASR accents at 24 ms · weakest: WildASR clipping at 105 ms.

How fast is Qwen3 ASR 1.7b?

On Baseten (dedicated inference), Qwen3 ASR 1.7b measures mean 39 ms time to final segment (1st of 29) and mean 946 ms time to first token (2nd of 27). Last measured 2026-10-16.

How accurate is Qwen3 ASR 1.7b?

On Baseten (dedicated inference), Qwen3 ASR 1.7b measures 2.9% word error rate (5th of 31). Last measured 2026-10-16.

Who hosts Qwen3 ASR 1.7b?

Qwen3 ASR 1.7b is created by Alibaba and served by Baseten. Coval measures each hosted endpoint separately.

Limits of this comparison

Coval measures it as a streaming WebSocket deployment on a Baseten dedicated endpoint, so latency reflects Baseten's serving configuration.

  • The endpoint is dedicated inference, so its latency is not ranked against shared APIs here. GPU choice, concurrency and decoding settings change speed and accuracy for an open-weight model.

Official sources

Results are re-measured daily using fixed datasets and reported over a rolling 30-day window. 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