# Open ASR Leaderboard

Tests how accurately speech-to-text systems write down what people say in English recordings (meetings, earnings calls, audiobooks and more). Lower error means fewer wrong words.

Category: asr. Type: leaderboard. Status: active.
Metric: Average word error rate across English datasets (%, lower is better) (lower is better).
Source: Hugging Face Open ASR Leaderboard (hf-audio) — https://huggingface.co/spaces/hf-audio/open_asr_leaderboard, results as of 2026-11-13, retrieved 2026-11-17, license: not stated.

| Rank | Model | Average WER (%) | RTFx (speed) |
|---|---|---|---|
| 1 | zoom/scribe_v2.1_pro | 4.87 |  |
| 2 | OtterAI/speech-v3.1 | 5.21 |  |
| 3 | yoob/yoob-asr-pro-v1 | 5.33 |  |
| 4 | microsoft/azure-speech-09-2026 | 5.5 |  |
| 5 | Musci-research/Musci-ASR-qwen-1019-hf | 5.7 | 774.69 |
| 6 | p99lab/audrey-0.7 | 5.7 |  |
| 7 | elevenlabs/scribe_v2 | 5.93 |  |
| 8 | reson8/rumble-2 | 5.95 |  |
| 9 | modulate/multilingual | 5.96 |  |
| 10 | salesforce-sovereign-intelligence/SOV-Transcribe-1 | 5.96 | 600.92 |
| 11 | FluidVoice-Labs/fluid-speech-pro | 6.06 | 3690.47 |
| 12 | sophea/asr-k1 | 6.11 |  |
| 13 | nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B | 6.13 | 472.62 |
| 14 | Qwen/Qwen3-ASR-1.7B-hf | 6.15 | 835.62 |
| 15 | assemblyai/universal-3-5-pro | 6.21 |  |
| 16 | smallestai/pulse | 6.57 |  |
| 17 | ibm-granite/granite-speech-5.0-2b | 6.68 | 548.43 |
| 18 | CohereLabs/cohere-transcribe-03-2026 | 6.72 | 923.71 |
| 19 | HojoAI/Hojo-ASR-V1 | 6.74 | 73.39 |
| 20 | TheStageAI/thewhisper-large-v3-turbo | 6.74 | 1183.92 |
| 21 | Edge0/ARK-ASR-0.6B | 6.75 | 664.56 |
| 22 | openai/gpt-transcribe | 6.8 |  |
| 23 | sprag/symphony | 6.85 |  |
| 24 | gladia/solaria-3 | 6.86 |  |
| 25 | nvidia/parakeet-tdt-0.6b-v2 (fast-gpu-asr) | 6.91 | 17254.19 |

Notes: English tab, top 25 of 69. Mix of open-weight and proprietary models; RTFx blank where not reported. Multilingual and long-form tabs exist.
