Open ASR Leaderboard
How the test works
Task
Transcribe speech and compare it with reference transcripts
- Kinds of tasksSpeech recognition
- In → outAudio → Text transcript
Run
Task macro-average across datasets
Score
Word error rate (WER)
- Graded byWhisper-style normalization: punctuation/case, numbers, spelling and filler handling
- Better isLower
- Test setDataset-specific test splits
Open ASR · AMI Cleaned leaderboardtop 10 of 61
Full leaderboard with filters →| Model | Open ASR · AMI Cleaned | Languages | Per audio minute | Price covers | Price | Released |
|---|---|---|---|---|---|---|
6.1 | — | $0.003 / min | Exact native tariff; see source for billing scope. | $0.003 / minute | ||
6.3 | — | — | — | |||
6.3 | en, es, fr +53 | $0.001 / min | batch audio transcription; optional enrichment add-ons billed separately | $0.001 / minute | ||
6.4 | — | — | — | |||
6.9 | — | — | — | |||
7 | — | — | — | |||
7 | en, es, fr +2 | — | — | Mar 2026 | ||
7 | en, es, fr +10 | — | — | Mar 2026 | ||
7.1 | — | — | — | |||
7.1 | en, es, fr +3 | — | — | Apr 2026 |
Versions and boards on Model Menu
| Board | Models |
|---|---|
Open ASR · AMI CleanedShown above | 61 |
| 61 | |
| 61 | |
| 61 | |
| 61 | |
| 61 | |
| 61 | |
| 61 |
Open ASR Leaderboard questions
What is Open ASR Leaderboard?›
Transcribe speech and compare it with reference transcripts. Created by Hugging Face and Open ASR Leaderboard collaborators.
Which model is best on Open ASR Leaderboard?›
Zoom Scribe V2 Pro leads Open ASR · AMI Cleaned with 6, ahead of ASR K1 (6) and Multilingual (6).
What is the best open-weights model on Open ASR Leaderboard?›
Granite Speech 4.1 2B Nar with 7, then Cohere Transcribe 03 2026 (7).
How is Open ASR Leaderboard scored?›
Word error rate (WER); whisper-style normalization: punctuation/case, numbers, spelling and filler handling.
Which versions of Open ASR Leaderboard are there?›
Open ASR · AMI Cleaned (61 models), Open ASR (61 models), Open ASR · Earnings-22 Cleaned AA Chunked (61 models), Open ASR · GigaSpeech Cleaned (61 models), Open ASR · LibriSpeech Clean (61 models), Open ASR · LibriSpeech Other (61 models), Open ASR · SPGISpeech (61 models), Open ASR · Voice Arena Monsoon (61 models).