whatmodel.sh / stt / Scribe v2
Scribe v2
↗ share
+ contribute
Scribe v2 is a closed-weights speech-to-text model from ElevenLabs, released January 2026. Licence: proprietary.
Score within the Open ASR English short-form WER (March 2026) comparison set only.
$ how it compares
The speech-to-text models ranked either side of Scribe v2, out of 176 with a score, then more from ElevenLabs.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 10 |  Soniox v4 Real-Time | 67±3 | comparison only | ||
| 11 |  Gemini 3.5 Transcribe | 67±6 | api $0.0050/audio min | ||
| 12 | Scribe v2 | 66±1 | api $0.0037/audio min | ||
| 13 |  GPT Transcribe | 64±4 | api $0.0045/audio min | ||
| 14 |  Zoom Scribe v1 | 64±5 | api price unknown | ||
| more from ElevenLabs | |||||
| 30 | Scribe v1 | 61±2 | comparison only | ||
| 81 | Scribe v2 Realtime | 55±3 | api $0.0065/audio min | ||
see all speech-to-text models · find the ones that fit your machine
$ what it's good at
long-form transcription▁▁▄▁·▁▂█▄▁▁▃ #1 of 27 * · 6.9% of its score · 1 test
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Open ASR English long-form WER | 6.2% ✎ | ▁▁▅▁·▃▅█▆▁▂▃ beats 29/30 | 6.9% |
multilingual transcription▂··▃▄█▆▅▄▂·▃ #1 of 36 · 34% of its score · 4 tests
Beats Universal-3.5 Pro here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Open ASR multilingual WER | 2.7% ✎ | ▂··▂▃▂█▃█▂▁▃ beats 29/29 | 23% |
| CoSE-E enterprise code-switching · 4 European pairs mean WER | 3.0% ✎ | █··██···█··█ beats 4/4 | 6.5% |
| mu-bench voice-agent caller utterances · batch UER (5 locales) | 17.3% ✎ | █·····▄····▄ beats 2/3 | 2.9% |
| Iberian ASR benchmark · 7-language macro WER (eu, ca, gl, pt, es + de, tr) | 6.5% ✎ | █▃▃··▅▃▃··▃▃ beats 10/10 | 1.2% |
other tests16 not in a task · 59% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Open ASR English short-form WER (March 2026) | 5.8% ✎ | ▁▁▁▃··▁▁▂██▆ beats 80/86 | 6.9% |
| RW-Voice-EQ ASR · 4 real-world English sets, mean WER | 6.7% ✎ | █▃▃▅▃···▃··▃ beats 9/9 | 5.4% |
| Dutch ASR Leaderboard · average WER (FLEURS, VoxPopuli, MLS nl) | 6.9% ✎ | ▄··▄█···█··▄ beats 6/6 | 5.0% |
| Persian ASR Triple Threat · VisualEars6669 + FLEURS-fa equal-weight WER | 4.1% ✎ | █·········██ beats 2/2 | 4.7% |
| VoiceCodeBench · batch canonical token/entity match (CTEM) | 91.6% ✎ | ▄···▄··▄▄▄·█ beats 6/6 | 4.1% |
| GigaSpeechBench · 7 Arabic dialect WER mean | 41.1% ✎ | █··█·██··█·█ beats 4/5 | 4.0% |
| VoiceCodeBench · batch WER (300 English workplace segments) | 6.9% ✎ | ▄·▄·▄·▄█···▄ beats 2/6 | 3.8% |
| Argmax OpenBench · earnings22-keywords WER (no keyword prompting) | 10.5% ✎ | █··········▄ beats 2/2 | 3.5% |
8 more tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| GigaSpeechBench · 12 English domains entity B-WER mean | 16.2% ✎ | ▄·······▄·█▄ beats 1/4 | 3.4% |
| Perle code-switching benchmark · 4-pair mean WER | 13.2% ✎ | █··········▄ beats 2/2 | 3.2% |
| BOTNOI report · FLEURS Thai CER | 6.2% ✎ | █··█·······█ beats 2/2 | 3.1% |
| GigaSpeechBench · 6 English accents WER mean | 21.9% ✎ | ▄······▄··▄█ beats 1/4 | 2.9% |
| Danish ASR Leaderboard · mean WER over 5 Danish test sets (RyeAI) | 13.3% ✎ | ▂▂···▃▂▅█▃▂▃ beats 16/16 | 2.6% |
| BOTNOI Thai telephony · CER | 19.7% ✎ | █·······█··█ beats 1/2 | 2.6% |
| koedesk FLEURS first-200 · 9-language headline error mean | 3.9% ✎ | ▄·····▄··▄█▄ beats 4/5 | 2.3% |
| Quranic ASR Leaderboard · held-out recitation (600 clips) overall WER | 12.5% ✎ | ▃·▅·▅▃▃·▃▃▃█ beats 9/12 | 1.2% |
$ where to run it
Through an API
api · $0.0037/audio min · elevenlabs.io ↗ · price sourceUSD per input audio minute, converted from the rounded LiteLLM per-second rate. File transcription; published tariff $0.22/hour. Not Scribe Realtime or Medical. Product: Batch/file speech transcription. API model: scribe_v2. Settings: streaming=no, minimum billed seconds=not specified. Price basis: rounded catalogue rate (not a new estimate of speech length). Currency: USD.
$ how the score was worked out · Open ASR Leaderboard 37%, GigaSpeechBench: A Real-World… 10%, VoiceCodeBench (arXiv:2608.28… 8% and 12 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json