Open ASR Leaderboard (Long-form) open-asr-longform Leaderboard
Long-form ASR β Average WER across longer-form datasets (Earnings21, Earnings22, Tedlium, CORAAL). Tests model behavior on multi-minute audio with topic shifts. Β· Metric: Average WER (lower is better)
| # | Model | Average WER | Paper |
|---|---|---|---|
| 1 | elevenlabs/scribe_v2 | 9.05 | link |
| 2 | assembly/universal-3-pro | 10.35 | link |
| 3 | reson8/resonant-1-flash | 10.46 | link |
| 4 | reson8/resonant-1 | 10.56 | link |
| 5 | speechmatics/enhanced | 10.98 | link |
| 6 | revai/fusion | 11.88 | link |
| 7 | revai/machine | 11.88 | link |
| 8 | CohereLabs/cohere-transcribe-03-2026 | 12.23 | link |
| 9 | nvidia/parakeet-tdt-0.6b-v3 | 13.37 | link |
| 10 | openai/whisper-large-v3-turbo | 13.60 | link |
| 11 | nvidia/parakeet-tdt-0.6b-v2 | 13.88 | link |
| 12 | openai/whisper-large-v3 | 13.92 | link |
| 13 | distil/whisper-distil-large-v3.5 | 14.08 | link |
| 14 | nvidia/canary-qwen-2.5b | 14.08 | link |
| 15 | distil/whisper-distil-large-v3 | 14.89 | link |
| 16 | openai/whisper-large | 15.23 | link |
| 17 | google/chirp | 15.55 | link |
| 18 | openai/whisper-large-v2 | 15.96 | link |
| 19 | nvidia/parakeet-ctc-1.1b | 16.17 | link |
| 20 | nvidia/parakeet-ctc-0.6b | 17.06 | link |
| 21 | nvidia/stt_en_conformer_transducer_small | 17.88 | link |
| 22 | nvidia/stt_en_conformer_ctc_large | 18.02 | link |
| 23 | nvidia/parakeet-rnnt-0.6b | 18.26 | link |
| 24 | nvidia/parakeet-tdt-1.1b | 19.75 | link |
| 25 | distil/whisper-distil-large-v2 | 20.24 | link |
| 26 | nvidia/parakeet-rnnt-1.1b | 21.37 | link |
| 27 | nvidia/stt_en_conformer_ctc_small | 22.88 | link |
| 28 | distil/whisper-distil-medium.en | 26.13 | link |
| 29 | nvidia/stt_en_fastconformer_ctc_large | 26.94 | link |
| 30 | nvidia/stt_en_conformer_transducer_large | 27.48 | link |