ZOOMSPEECH-TO-TEXT85,368 SAMPLES / 30 DAYSLAST RUN NOV 14, 2026, 01:30 UTC
Scribe speech-to-text benchmarks
Scribe, hosted by Zoom, measures mean 112 ms time to final segment (15th of 30) and 2.6% word error rate (8th of 31) among STT systems. Results cover the last 30 days. Coval benchmarks hosted endpoints daily using a consistent evaluation methodology. Last measured .
Coval has active speech-to-text measurements for Scribe, created by Zoom and served through Zoom's own API.
- Time to Final Segment#15 / 30
- 112ms
- Word Error Rate#8 / 31
- 2.6%
Overview
Scribe 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.
How Scribe ranks
Full STT dashboard- #2Qwen3 ASR 1.7bDedicated inference. Shared endpoints serve many customers on the same infrastructure, while dedicated endpoints run on hardware reserved for a single customer.35 ms
Show all 30 modelsShow fewer
- #17Whisper Large v3Dedicated inference. Shared endpoints serve many customers on the same infrastructure, while dedicated endpoints run on hardware reserved for a single customer.125 ms
- #9Qwen3 ASR 1.7bDedicated inference. Shared endpoints serve many customers on the same infrastructure, while dedicated endpoints run on hardware reserved for a single customer.2.7%
Show all 31 modelsShow fewer
- #20Whisper Large v3Dedicated inference. Shared endpoints serve many customers on the same infrastructure, while dedicated endpoints run on hardware reserved for a single customer.4.1%
- #2Whisper Large v3Dedicated inference. Shared endpoints serve many customers on the same infrastructure, while dedicated endpoints run on hardware reserved for a single customer.919 ms
- #3Qwen3 ASR 1.7bDedicated inference. Shared endpoints serve many customers on the same infrastructure, while dedicated endpoints run on hardware reserved for a single customer.941 ms
Show all 27 modelsShow fewer
| # | Model | Host | TTFS | WER | TTFT | Samples |
|---|---|---|---|---|---|---|
| 1 | Saidwire En 2026 10 30 | Saidwire | 29 ms | 768 ms | 9,460 | |
| 2 | Qwen3 ASR 1.7b | Baseten | 35 ms | 941 ms | 13,535 | |
| 3 | STT RT v5 | Soniox | 51 ms | 1529 ms | 27,360 | |
| 4 | Qwen3 ASR Fast | Nari Labs | 53 ms | 1757 ms | 27,363 | |
| 5 | STT Rt v6 | Soniox | 59 ms | 1893 ms | 12,207 | |
| 6 | STT 1 | Inworld AI | 70 ms | 994 ms | 27,336 | |
| 7 | Nova 3 | Deepgram | 95 ms | 1435 ms | 24,082 | |
| 8 | Nova 2 | Deepgram | 97 ms | 1432 ms | 24,047 | |
| 9 | Realtime Global | Reson8 | 98 ms | 1658 ms | 27,363 | |
| 10 | Scribe v3 Live | ElevenLabs | 98 ms | 1254 ms | 5,800 | |
| 11 | Universal 3.5 Pro | AssemblyAI | 104 ms | 1013 ms | 27,363 | |
| 12 | Universal 3.6 Pro | AssemblyAI | 110 ms | 1073 ms | 27,362 | |
| 13 | Flux Multilingual | Deepgram | 110 ms | 1173 ms | 24,080 | |
| 14 | Flux | Deepgram | 112 ms | 1095 ms | 24,086 | |
| 15 | Scribe | Zoom | 112 ms | — | 6,654 | |
| 16 | Scribe v2 Realtime | ElevenLabs | 116 ms | 2126 ms | 27,359 | |
| 17 | Whisper Large v3 | Baseten | 125 ms | 919 ms | 13,079 | |
| 18 | Ink 2 | Cartesia | 141 ms | 2030 ms | 18,414 | |
| 19 | Grok STT | SpaceXAI | 163 ms | — | 27,345 | |
| 20 | Pulse | Smallest | 168 ms | 1953 ms | 27,360 | |
| 21 | resonant-1 | Reson8 | 210 ms | — | 27,355 | |
| 22 | Linden 1 | Speechmatics | 229 ms | 1503 ms | 27,339 | |
| 23 | Scribe Pro v2.1 | Zoom | 241 ms | — | 6,823 | |
| 24 | Default | Gradium | 284 ms | 2074 ms | 27,355 | |
| 25 | Daytona STT | Guava | 347 ms | 5606 ms | 13,249 | |
| 26 | Gemini 3.5 Transcribe Live | Gemini | 355 ms | 1520 ms | 27,283 | |
| 27 | Solaria 1 | Gladia | 390 ms | 1535 ms | 27,320 | |
| 28 | GPT Realtime Whisper | OpenAI | 566 ms | 1813 ms | 26,880 | |
| 29 | Chirp 3 | 842 ms | 6035 ms | 27,313 | ||
| 30 | Chirp 2 | 909 ms | 6101 ms | 27,361 | ||
| — | Universal Streaming | AssemblyAI | — | 1511 ms | 27,332 |
Highest relative placement: 8th of 31 on Word Error Rate.
Latency vs accuracy
Where the errors come from
- Scribe2.6%
Averages and tail latency
Averages hide slow outliers — these are the distributions behind each figure.
- Time to Final Segmentp50 110 ms · Max 363 ms
- Scribep50 0.0% · Max 66.7%
| Metric | Average | p25 | p50 | p75 | p90 | p95 | Max | Samples |
|---|---|---|---|---|---|---|---|---|
| Time to Final Segment | 112 ms | 73 ms | 110 ms | 150 ms | 174 ms | 187 ms | 363 ms | 6,654 |
| Word Error Rate | 2.6% | 0.0% | 0.0% | 3.4% | 9.5% | 13.6% | 66.7% | 24,543 |
Last 30 days
Daily medians from the same measurement runs · gaps are days without qualifying runs.
Time to Final Segment by dataset
- WildASR clean99 ms
- WildASR clipping101 ms
- WildASR far-field102 ms
- WildASR noise gaps103 ms
- WildASR phone codec104 ms
- PipeCat (production)116 ms
- WildASR accents169 ms
| Dataset | TTFS | Samples |
|---|---|---|
| ProductionPipeCat | 116 ms | 3,499 |
| AccentsWildASR | 169 ms | 351 |
| CleanWildASR | 99 ms | 1,404 |
| ClippingWildASR | 101 ms | 349 |
| Far-fieldWildASR | 102 ms | 351 |
| Noise gapsWildASR | 103 ms | 351 |
| Phone codecWildASR | 104 ms | 349 |
Strongest condition: WildASR clean at 99 ms · weakest: WildASR accents at 169 ms.
How fast is Scribe?
On Zoom, Scribe measures mean 112 ms time to final segment (15th of 30). Last measured 2026-11-14.
How accurate is Scribe?
On Zoom, Scribe measures 2.6% word error rate (8th of 31). Last measured 2026-11-14.
Who hosts Scribe?
Scribe is created by Zoom and served by Zoom. Coval measures each hosted endpoint separately.
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.