Hugging Face Open ASR Leaderboard
Eight-set macro-average English WER, lower is better; mostly read and broadcast English
speechcrowded
Score in WER, snapshot 2026-05-06 via benchmarklist
Definition
- Version
- not in the notes
- Category
- speech
- Maintainer
- Hugging Face
- Status
- crowded
- Home page
- https://benchmarklist.com/benchmarks/open_asr_leaderboard/
Sources
- Open ASR Leaderboard Benchmark Scores & AI Model Leaderboard | BenchmarkList benchmarklist.com https://benchmarklist.com/benchmarks/open_asr_leaderboard/
- Gladia - Best open-source speech-to-text models in 2026 gladia.io https://www.gladia.io/blog/best-open-source-speech-to-text-models
- huggingface.co/spaces/hf-audio/open_asr_leaderboard https://huggingface.co/spaces/hf-audio/open_asr_leaderboard
Scores at a Glance
Score in WER, snapshot 2026-05-06 via benchmarklist
Lower is better for this unit.
1
granite-speech-4.1-2bIBMeight-set macro-average English WER
Cohere Transcribe (cohere-transcribe-03-2026)Cohereeight-set macro-average English WERgranite-4.0-1b-speechIBMeight-set macro-average English WER
Canary-Qwen-2.5BNVIDIAeight-set macro-average English WERgranite-speech-3.3-8bIBMeight-set macro-average English WER
Qwen3-ASR-1.7BAlibaba (Qwen)eight-set macro-average English WER
Phi-4-multimodal-instructMicrosoft AIeight-set macro-average English WERParakeet-TDT-0.6B v2NVIDIAeight-set macro-average English WERParakeet-TDT-0.6B v3NVIDIAeight-set macro-average English WERCanary-1B-flashNVIDIAeight-set macro-average English WERQwen3-ASR-0.6BAlibaba (Qwen)eight-set macro-average English WERCanary-1BNVIDIAeight-set macro-average English WER
granite-speech-4.1-2bIBMeight-set macro-average English WER5.33 WER
maintainer-run2Cohere Transcribe (cohere-transcribe-03-2026)Cohereeight-set macro-average English WER5.42 WER
maintainer-run3granite-4.0-1b-speechIBMeight-set macro-average English WER5.52 WER
maintainer-run4Canary-Qwen-2.5BNVIDIAeight-set macro-average English WER5.63 WER
maintainer-run5granite-speech-3.3-8bIBMeight-set macro-average English WER5.74 WER
maintainer-run6Qwen3-ASR-1.7BAlibaba (Qwen)eight-set macro-average English WER5.76 WER
maintainer-run7Phi-4-multimodal-instructMicrosoft AIeight-set macro-average English WER6.02 WER
maintainer-run8Parakeet-TDT-0.6B v2NVIDIAeight-set macro-average English WER6.05 WER
maintainer-run9Parakeet-TDT-0.6B v3NVIDIAeight-set macro-average English WER6.32 WER
maintainer-run10Canary-1B-flashNVIDIAeight-set macro-average English WER6.35 WER
maintainer-run11Qwen3-ASR-0.6BAlibaba (Qwen)eight-set macro-average English WER6.42 WER
maintainer-run12Canary-1BNVIDIAeight-set macro-average English WER6.5 WER
maintainer-run02468
Score in WER
Lower is better for this unit.
1
Voxtral SmallMistral AIaverage WER
Whisper large-v3OpenAIaverage WER per Gladia, from the same leaderboardWhisper large-v3-turboOpenAIaverage WER per Gladia, from the same leaderboard
Voxtral SmallMistral AIaverage WER6.62 WER
maintainer-run2Whisper large-v3OpenAIaverage WER per Gladia, from the same leaderboard7.4 WER
maintainer-run3Whisper large-v3-turboOpenAIaverage WER per Gladia, from the same leaderboard7.75 WER
maintainer-run02468
Where this comes from
Each bar is one model, with its most independent recorded score. Hatched bars are vendor-reported; the word beside each bar says who ran it. Settings and dates differ between rows; the Scores table below has every result.
Scores
| Model | Score | Who ran it | Setting | View or population | Date | Source |
|---|---|---|---|---|---|---|
| Whisper large-v3-turbo | 7.75 WER | maintainer-run | average WER per Gladia, from the same leaderboard | – | – | gladia.io |
| Whisper large-v3 | 7.44 WER | maintainer-run | leadership history | leader in 2023 | 2023 | select.keenable.ai |
| Whisper large-v3 | 7.4 WER | maintainer-run | average WER per Gladia, from the same leaderboard | – | – | gladia.io |
| Voxtral Small | 6.62 WER | maintainer-run | average WER | – | – | select.keenable.ai |
| Canary-1B | 6.5 WER | maintainer-run | eight-set macro-average English WER | snapshot 2026-05-06 via benchmarklist | 6 May 2026 | benchmarklist.com |
| Qwen3-ASR-0.6B | 6.42 WER | maintainer-run | eight-set macro-average English WER | snapshot 2026-05-06 via benchmarklist | 6 May 2026 | benchmarklist.com |
| Canary-1B-flash | 6.35 WER | maintainer-run | eight-set macro-average English WER | snapshot 2026-05-06 via benchmarklist | 6 May 2026 | benchmarklist.com |
| Parakeet-TDT-0.6B v3 | 6.32 WER | maintainer-run | eight-set macro-average English WER | snapshot 2026-05-06 via benchmarklist | 6 May 2026 | benchmarklist.com |
| Parakeet-TDT-0.6B v2 | 6.05 WER | maintainer-run | eight-set macro-average English WER | snapshot 2026-05-06 via benchmarklist | 6 May 2026 | benchmarklist.com |
| Phi-4-multimodal-instruct | 6.02 WER | maintainer-run | eight-set macro-average English WER | snapshot 2026-05-06 via benchmarklist | 6 May 2026 | benchmarklist.com |
| Qwen3-ASR-1.7B | 5.76 WER | maintainer-run | eight-set macro-average English WER | snapshot 2026-05-06 via benchmarklist | 6 May 2026 | benchmarklist.com |
| granite-speech-3.3-8b | 5.74 WER | maintainer-run | eight-set macro-average English WER | snapshot 2026-05-06 via benchmarklist | 6 May 2026 | benchmarklist.com |
| Canary-Qwen-2.5B | 5.63 WER | maintainer-run | eight-set macro-average English WER | snapshot 2026-05-06 via benchmarklist | 6 May 2026 | benchmarklist.com |
| granite-4.0-1b-speech | 5.52 WER | maintainer-run | eight-set macro-average English WER | snapshot 2026-05-06 via benchmarklist | 6 May 2026 | benchmarklist.com |
| Cohere Transcribe (cohere-transcribe-03-2026) | 5.42 WER | maintainer-run | eight-set macro-average English WER | rank 2, snapshot 2026-05-06 via benchmarklist | 6 May 2026 | benchmarklist.com |
| granite-speech-4.1-2b | 5.33 WER | maintainer-run | eight-set macro-average English WER | rank 1, snapshot 2026-05-06 via benchmarklist | 6 May 2026 | benchmarklist.com |
Where this comes from
Scores from different settings, harnesses or dates are not directly comparable; the row says which. Sorted by score, highest first, as a reading aid, not a ranking.
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Hugging Face Open ASR Leaderboard", https://aurion.semai.id/benchmarks/open-asr-leaderboard/, records of October 2026.