Open ASR Leaderboard (English) open-asr-english Leaderboard
Open ASR Leaderboard's main English benchmark — Average WER across 8 standard test sets (AMI, Earnings22, Gigaspeech, LibriSpeech Clean/Other, SPGISpeech, TED-LIUM, Voxpopuli). Lower is better. The community standard for ASR model comparison. · Metric: Average WER (lower is better)
| # | Model | Average WER | Paper |
|---|---|---|---|
| 1 | microsoft/azure-speech-06-2026 | 4.84 | link |
| 2 | AutoArk-AI/ARK-ASR-3B | 5.04 | link |
| 3 | bosonai/higgs-audio-v3-stt | 5.04 | link |
| 4 | zoom/scribe_v1 | 5.12 | link |
| 5 | ibm-granite/granite-speech-4.1-2b | 5.18 | link |
| 6 | bosonai/higgs-audio-v3-8b-stt-v2 | 5.25 | link |
| 7 | ibm-granite/granite-speech-4.1-2b-nar | 5.25 | link |
| 8 | CohereLabs/cohere-transcribe-03-2026 | 5.35 | link |
| 9 | ibm-granite/granite-4.0-1b-speech | 5.37 | link |
| 10 | reson8/resonant-1 | 5.38 | link |
| 11 | reson8/resonant-1-flash | 5.39 | link |
| 12 | nvidia/canary-qwen-2.5b | 5.41 | link |
| 13 | ibm-granite/granite-speech-3.3-8b | 5.54 | link |
| 14 | Qwen/Qwen3-ASR-1.7B | 5.59 | link |
| 15 | ibm-granite/granite-speech-3.3-2b | 5.79 | link |
| 16 | AutoArk-AI/ARK-ASR-0.6B | 5.82 | link |
| 17 | nvidia/parakeet-tdt-0.6b-v2 | 5.86 | link |
| 18 | smallestai/pulse | 5.87 | link |
| 19 | microsoft/Phi-4-multimodal-instruct | 5.89 | link |
| 20 | aquavoice/avalon-v1-en | 6.02 | link |
| 21 | assemblyai/universal-3-pro | 6.06 | link |
| 22 | nvidia/canary-1b-flash | 6.18 | link |
| 23 | nvidia/parakeet-tdt-0.6b-v3 | 6.22 | link |
| 24 | nvidia/canary-1b | 6.24 | link |
| 25 | kyutai/stt-2.6b-en | 6.31 | link |
| 26 | Qwen/Qwen3-ASR-0.6B | 6.31 | link |
| 27 | mistralai/Voxtral-Small-24B-2507 | 6.47 | link |
| 28 | usefulsensors/moonshine-streaming-medium | 6.55 | link |
| 29 | nyrahealth/CrisperWhisper | 6.56 | link |
| 30 | soundsgoodai/Zipformer-transducer-XL-290M | 6.71 | link |