Voice AI BenchmarksBy

ASSEMBLYAISPEECH-TO-TEXT106,496 SAMPLES / 30 DAYSLAST RUN OCT 14, 2026, 06:00 UTC

official resource

Universal Streaming speech-to-text benchmarks

Universal Streaming, hosted by AssemblyAI (Assembly AI, assemblyai.com), measures 5.9% word error rate (23rd of 31) among STT systems. Results cover the last 30 days. Coval benchmarks hosted endpoints daily using a consistent evaluation methodology. Each hosted endpoint is measured separately under the same benchmark conditions. Last measured .

Universal Streaming is AssemblyAI's real-time speech recognition model.

Word Error Rate#23 / 31
5.9%
Time to First Token#12 / 27
1508ms

Overview

AssemblyAI provides it through a persistent real-time API with voice activity, diarization and keyterm support.

Universal Streaming 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
AssemblyAI
Hosted by
AssemblyAI
Source
Official API
Licensing
Proprietary
Deployment
On-prem
Region
US
Features
Diarization, Keyterm biasing, VAD

How Universal Streaming 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 Universal Streaming highlighted.
  1. #1Qwen3 ASR 1.7b39 ms
  2. #2STT RT v545 ms
  3. #3Qwen3 ASR Fast47 ms
  4. #4STT 166 ms
  5. #5Nova 385 ms
  6. #6Nova 287 ms
  7. #7Parakeet TDT 0.6B v397 ms
  8. #8Flux Multilingual98 ms
  9. #9Flux99 ms
  10. #10Scribe v2 Realtime122 ms
  11. #11Whisper Large v3via Baseten131 ms
  12. #12Universal 3.6 Pro135 ms
Show all 29 models
  1. #13Ink 2148 ms
  2. #14Universal 3.5 Pro168 ms
  3. #15Whisper Large v3via Together AI200 ms
  4. #16Pulse205 ms
  5. #17Grok STT223 ms
  6. #18resonant-1252 ms
  7. #19Default258 ms
  8. #20Linden 1271 ms
  9. #21Gemini 3.5 Transcribe Live317 ms
  10. #22Daytona STT399 ms
  11. #23Voxtral Mini Transcribe Realtime 2602410 ms
  12. #24GPT Realtime Whisper599 ms
  13. #25Solaria 1636 ms
  14. #26GPT-4o mini Transcribe708 ms
  15. #27Chirp 3768 ms
  16. #28GPT-4o Transcribe783 ms
  17. #29Chirp 2908 ms
median of all models · 200 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 ms1518 ms23,533
3Qwen3 ASR FastNari Labs47 ms1752 ms26,652
4STT 1Inworld AI66 ms1211 ms24,749
5Nova 3Deepgram85 ms1431 ms26,002
6Nova 2Deepgram87 ms1417 ms25,960
7Parakeet TDT 0.6B v3Together AI97 ms1228 ms26,461
8Flux MultilingualDeepgram98 ms1165 ms26,007
9FluxDeepgram99 ms1091 ms26,005
10Scribe v2 RealtimeElevenLabs122 ms2158 ms26,640
11Whisper Large v3Baseten131 ms924 ms13,663
12Universal 3.6 ProAssemblyAI135 ms1072 ms24,204
13Ink 2Cartesia148 ms1966 ms26,652
14Universal 3.5 ProAssemblyAI168 ms1025 ms26,654
15Whisper Large v3Together AI200 ms1261 ms26,445
16PulseSmallest205 ms1976 ms26,654
17Grok STTSpaceXAI223 ms—26,576
18resonant-1Reson8252 ms—26,653
19DefaultGradium258 ms1986 ms26,632
20Linden 1Speechmatics271 ms1537 ms26,595
21Gemini 3.5 Transcribe LiveGemini317 ms1609 ms26,567
22Daytona STTGuava399 ms5695 ms5,225
23Voxtral Mini Transcribe Realtime 2602Mistral410 ms1851 ms26,200
24GPT Realtime WhisperOpenAI599 ms1824 ms26,622
25Solaria 1Gladia636 ms1781 ms26,200
26GPT-4o mini TranscribeOpenAI708 ms—26,621
27Chirp 3Google768 ms5956 ms26,638
28GPT-4o TranscribeOpenAI783 ms—26,625
29Chirp 2Google908 ms6098 ms26,650
—Nemotron 3.5 ASR StreamingTogether AI—1544 ms26,542
—Universal StreamingAssemblyAI—1508 ms26,624
Under-sampled models are excluded; tied models share a place. Dotted WER values split into substitutions, deletions and insertions on hover or tap.

Universal Streaming hasn't logged enough qualifying samples in the last 30 days to hold a rank on Time to Final Segment.

Where the errors come from

WER compositionUniversal Streaming's Word Error Rate split by error type · 30-day averageUniversal Streaming's WER split into substitutions, deletions and insertions.
  • Universal Streaming5.9%
SubstitutionsDeletionsInsertions

Averages and tail latency

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

Universal Streaming latency distributionMilliseconds · shared axis across metrics · last 30 daysUniversal Streaming's latency percentiles per metric: p25–p75 band, p50 tick, whisker to max.
  • Time to First Tokenp50 1480 ms · Max 10946 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
Word Error Rate5.9%0.0%3.2%10.0%17.6%25.0%81.8%26,624
Time to First Token1508 ms1176 ms1480 ms1731 ms2083 ms2381 ms10946 ms26,624

How fast is Universal Streaming?

On AssemblyAI, Universal Streaming measures mean 1508 ms time to first token (12th of 27). Last measured 2026-10-14.

How accurate is Universal Streaming?

On AssemblyAI, Universal Streaming measures 5.9% word error rate (23rd of 31). Last measured 2026-10-14.

Who hosts Universal Streaming?

Universal Streaming is created by AssemblyAI and served by AssemblyAI. Coval measures each hosted endpoint separately.

Limits of this comparison

Coval reports partial-transcript timing, final-transcript timing and recognition accuracy.

  • Production endpointing and turn detection can add latency outside the interval Coval attributes to the 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