Open model
Whisper
by OpenAI
An open-source speech recognition model from OpenAI that transcribes and translates speech in many languages and can be run locally.
- Self-hosted
Key figures
Every figure about Whisper
Each figure is a row of a research dataset. Open it for the exact words of its source or the exact request it was read from.
On-Device Speech Recognition Benchmark: 25 Languages
- Whisper large-v3-turbo, Bulgarian2026-10-1915.87 WER %
- Whisper large-v3-turbo, Czech2026-10-1911.49 WER %
- Whisper large-v3-turbo, Danish2026-10-1914.57 WER %
- Whisper large-v3-turbo, German2026-10-195.08 WER %
- Whisper large-v3-turbo, Greek2026-10-1911.96 WER %
- Whisper large-v3-turbo, English2026-10-193.92 WER %
- Whisper large-v3-turbo, Spanish2026-10-192.53 WER %
- Whisper large-v3-turbo, Estonian2026-10-1919.91 WER %
- Whisper large-v3-turbo, Finnish2026-10-197.73 WER %
- Whisper large-v3-turbo, French2026-10-196.62 WER %
- Whisper large-v3-turbo, Croatian2026-10-1911.34 WER %
- Whisper large-v3-turbo, Hungarian2026-10-1917.21 WER %
- Whisper large-v3-turbo, Italian2026-10-192.73 WER %
- Whisper large-v3-turbo, Lithuanian2026-10-1925.03 WER %
- Whisper large-v3-turbo, Latvian2026-10-1918.64 WER %
- Whisper large-v3-turbo, Maltese2026-10-1972.35 WER %
- Whisper large-v3-turbo, Dutch2026-10-196.59 WER %
- Whisper large-v3-turbo, Polish2026-10-197.7 WER %
- Whisper large-v3-turbo, Portuguese2026-10-193.28 WER %
- Whisper large-v3-turbo, Romanian2026-10-1911.79 WER %
- Whisper large-v3-turbo, Russian2026-10-194.59 WER %
- Whisper large-v3-turbo, Slovak2026-10-1911.51 WER %
- Whisper large-v3-turbo, Slovenian2026-10-1923.51 WER %
- Whisper large-v3-turbo, Swedish2026-10-199.95 WER %
Dictation Punctuation and Capitalization Accuracy
- Whisper large-v3-turbo, Bulgarian2026-10-1917.71 %
- Whisper large-v3-turbo, Czech2026-10-1913.3 %
- Whisper large-v3-turbo, Danish2026-10-1916.9 %
- Whisper large-v3-turbo, German2026-10-197.1 %
- Whisper large-v3-turbo, Greek2026-10-1916.1 %
- Whisper large-v3-turbo, English2026-10-197.71 %
- Whisper large-v3-turbo, Spanish2026-10-195.68 %
- Whisper large-v3-turbo, Estonian2026-10-1920.52 %
- Whisper large-v3-turbo, Finnish2026-10-199.88 %
- Whisper large-v3-turbo, French2026-10-199.28 %
- Whisper large-v3-turbo, Croatian2026-10-1914.21 %
- Whisper large-v3-turbo, Hungarian2026-10-1919.28 %
- Whisper large-v3-turbo, Italian2026-10-197.65 %
- Whisper large-v3-turbo, Lithuanian2026-10-1926.71 %
- Whisper large-v3-turbo, Latvian2026-10-1920.35 %
- Whisper large-v3-turbo, Maltese2026-10-1971.96 %
- Whisper large-v3-turbo, Dutch2026-10-198.99 %
- Whisper large-v3-turbo, Polish2026-10-1910.48 %
- Whisper large-v3-turbo, Portuguese2026-10-197.45 %
- Whisper large-v3-turbo, Romanian2026-10-1913.99 %
- Whisper large-v3-turbo, Russian2026-10-196.13 %
- Whisper large-v3-turbo, Slovak2026-10-1912.58 %
- Whisper large-v3-turbo, Slovenian2026-10-1925.26 %
- Whisper large-v3-turbo, Swedish2026-10-1912.94 %
Speech Recognition Accuracy: Word Error Rates
- Whisper Large v3 Replicate large-v3202610.1 WER %
- Whisper Large v3 Turbo, Telnyx Telnyx20265.5 WER %
- Whisper large-v3, zero-shot202423 WER %
- Whisper, whispered speech202418.8 WER %
- openai/whisper-large-v320267.84 WER %
- openai/whisper-large-v3-turbo20268.39 WER %
- Jacobs et al. 2025, isiXhosa202580.4 WER %
- Kuhn et al. 2024, Whisper large-v220242.9 WER %
- Kurmanji Whisper v3 2024202410.5 WER %
- Lin et al. 2025, Extra tags20255.81 WER %
- Whisper large-v3-turbo, Bulgarian (our test)202615.87 WER %
- Whisper large-v3-turbo, Czech (our test)202611.49 WER %
- Whisper large-v3-turbo, Danish (our test)202614.57 WER %
- Whisper large-v3-turbo, German (our test)20265.08 WER %
- Whisper large-v3-turbo, Greek (our test)202611.96 WER %
- Whisper large-v3-turbo, English (our test)20263.92 WER %
- Whisper large-v3-turbo, Spanish (our test)20262.53 WER %
- Whisper large-v3-turbo, Estonian (our test)202619.91 WER %
- Whisper large-v3-turbo, Finnish (our test)20267.73 WER %
- Whisper large-v3-turbo, French (our test)20266.62 WER %
- Whisper large-v3-turbo, Croatian (our test)202611.34 WER %
- Whisper large-v3-turbo, Hungarian (our test)202617.21 WER %
- Whisper large-v3-turbo, Italian (our test)20262.73 WER %
- Whisper large-v3-turbo, Lithuanian (our test)202625.03 WER %
Speech Recognition Speed: Real-Time Factor Data
- Whisper large-v3-turbo, Bulgarian (M2 Pro)202612 RTFx
- Whisper large-v3-turbo, Czech (M2 Pro)202615 RTFx
- Whisper large-v3-turbo, Danish (M2 Pro)202613 RTFx
- Whisper large-v3-turbo, German (M2 Pro)202616 RTFx
- Whisper large-v3-turbo, Greek (M2 Pro)202611 RTFx
- Whisper large-v3-turbo, English (M2 Pro)202612 RTFx
- Whisper large-v3-turbo, Spanish (M2 Pro)202615 RTFx
- Whisper large-v3-turbo, Estonian (M2 Pro)202615 RTFx
- Whisper large-v3-turbo, French (M2 Pro)202612 RTFx
- Whisper large-v3-turbo, Italian (M2 Pro)202617 RTFx
- Whisper large-v3-turbo, Latvian (M2 Pro)202615 RTFx
- Whisper large-v3-turbo, Maltese (M2 Pro)202613 RTFx
- Whisper large-v3-turbo, Dutch (M2 Pro)202612 RTFx
- Whisper large-v3-turbo, Polish (M2 Pro)202612 RTFx
- Whisper large-v3-turbo, Portuguese (M2 Pro)202616 RTFx
- Whisper large-v3-turbo, Romanian (M2 Pro)202612 RTFx
- Whisper large-v3-turbo, Slovak (M2 Pro)202614 RTFx
- Whisper large-v3-turbo, Slovenian (M2 Pro)202612 RTFx
Most Downloaded Speech Recognition Models (Hugging Face)
- openai/whisper-baseread October 19, 20261,676,272 downloads
- openai/whisper-large-v2read October 19, 202680,446 downloads
- openai/whisper-large-v3-turboread October 19, 20266,094,388 downloads
- openai/whisper-large-v3read October 19, 20263,755,126 downloads
- openai/whisper-medium.enread October 19, 202629,939 downloads
- openai/whisper-mediumread October 19, 2026313,832 downloads
- openai/whisper-smallread October 19, 20263,510,611 downloads
- openai/whisper-tinyread October 19, 2026818,386 downloads
Other open models and engines
- Breeze ASRMediaTek Research
- CMU Sphinx (PocketSphinx)Carnegie Mellon University
- Cohere TranscribeCohere
- Coqui STTCoqui
- CrisperWhisperNyra Health
- Distil-WhisperHugging Face
- ESPnetESPnet developers
- faster-whisperSYSTRAN
- FluidAudioFluid Inference
- FunASRModelScope
- GigaAMSaluteDevices (Sber)
- GLM-ASRZ.ai
- IBM Granite SpeechIBM
- IndicConformerAI4Bharat
- insanely-fast-whisperVaibhav Srivastav
- ivrit.ai Whisperivrit.ai
- JuliusJulius development team
- k2 and icefallNext-gen Kaldi team
- KaldiKaldi contributors
- KB-WhisperKBLab, National Library of Sweden
- Kotoba-WhisperKotoba Technologies
- Kyutai STTKyutai
- Lightning Whisper MLXMustafa Aljadery
- Meta MMSMeta AI
- MLX WhisperApple MLX team
- MoonshineMoonshine AI
- Mozilla DeepSpeechMozilla
- NB-WhisperNational Library of Norway AI Lab
- NVIDIA CanaryNVIDIA
- NVIDIA NeMoNVIDIA