Skip to content

Open ASR Leaderboard what it measures, how it is scored and who leads it

Speech · Hugging Face and collaborators · Average word error rate (lower is better)

Scores from Open ASR Leaderboard, published Sep 25, 2026

RankModelScore
1scribe v2 pro (zoom)Zoom · as “zoom/scribe_v2_pro”3.59 WER
2azure-speech-07-2026 (microsoft)Microsoft · as “microsoft/azure-speech-07-2026”3.81 WER
3multilingual (modulate)Modulate · as “modulate/multilingual”3.84 WER
4scribe v2 (elevenlabs)ElevenLabs · as “elevenlabs/scribe_v2”3.97 WER
5resonant-1 (reson8)Reson8 · as “reson8/resonant-1”4.01 WER
6resonant-1-flash (reson8)Reson8 · as “reson8/resonant-1-flash”4.02 WER
7scribe v1 (zoom)Zoom · as “zoom/scribe_v1”4.15 WER
8azure-speech-06-2026 (microsoft)Microsoft · as “microsoft/azure-speech-06-2026”4.26 WER
8asr-k1 (preview) (sophea)Sophea · as “sophea/asr-k1 (preview)”4.26 WER
10Qwen3-ASR-1.7B-hf (Qwen)Alibaba · as “Qwen/Qwen3-ASR-1.7B-hf”4.31 WER
11Hojo-ASR-V1 (HojoAI)Hojoai · as “HojoAI/Hojo-ASR-V1”4.33 WER
12universal-3-5-pro (assemblyai)AssemblyAI · as “assemblyai/universal-3-5-pro”4.34 WER
13symphony (sprag)Sprag · as “sprag/symphony”4.35 WER
14higgs-audio-v3-stt (bosonai)Bosonai · as “bosonai/higgs-audio-v3-stt”4.39 WER
15pulse (smallestai)Smallestai · as “smallestai/pulse”4.41 WER
16canary-qwen-2.5b (nvidia)Nvidia · as “nvidia/canary-qwen-2.5b”4.43 WER
17ARK-ASR-3B (AutoArk-AI)Autoark Ai · as “AutoArk-AI/ARK-ASR-3B”4.47 WER
18ARK-ASR-0.6B (AutoArk-AI)Autoark Ai · as “AutoArk-AI/ARK-ASR-0.6B”4.56 WER
19solaria-3 (gladia)Gladia · as “gladia/solaria-3”4.57 WER
20granite-speech-4.1-2b (ibm-granite)IBM · as “ibm-granite/granite-speech-4.1-2b”4.62 WER
21MOSS-Transcribe-Diarize (OpenMOSS-Team)Openmoss Team · as “OpenMOSS-Team/MOSS-Transcribe-Diarize”4.64 WER
22cohere-transcribe-03-2026 (CohereLabs)Cohere · as “CohereLabs/cohere-transcribe-03-2026”4.67 WER
23granite-speech-4.1-2b-nar (ibm-granite)IBM · as “ibm-granite/granite-speech-4.1-2b-nar”4.68 WER
23parakeet-tdt-0.6b-v2 (fast-gpu-asr) (nvidia)Nvidia · as “nvidia/parakeet-tdt-0.6b-v2 (fast-gpu-asr)”4.68 WER
25parakeet-tdt-0.6b-v2 (nvidia)Nvidia · as “nvidia/parakeet-tdt-0.6b-v2”4.70 WER

About this benchmark

Task
Transcribe English short- and long-form speech and multilingual speech.
Dataset
Public test sets (LibriSpeech, Common Voice, TED-LIUM, AMI, Earnings-22 and more) in English and multilingual tracks.
Method
Every system run the same way; word error rate and real-time factor reported.
Metric
Average word error rate (lower is better)
Organization
Hugging Face and collaborators

Versions

Newest first. New versions are added, never rewritten.

VersionDate
Paper86 systems across 12 datasets, with long-form and multilingual tracks.Oct 8, 202511 months ago

Sources: each benchmark's paper (arXiv) and its official site or leaderboard. Scores are the boards captured here every day, exactly as published.

New #1s by email

Saturdays, only in weeks when a leaderboard has a new #1.

Double opt-in. Unsubscribe any time.