Leaderboards
Benchmarks
Every score carries its origin — leaderboard is independent,vendor is self-reported. History is kept, never overwritten.
Accented speech74Clean read speech74European-accented speech74Financial calls74Harder read speech74Open ASR RTFx62Open ASR WER74Podcasts and video74Recorded meetings74
Accented speech
Word error rate on Earnings-22 — earnings calls given by speakers from many countries, which is the closest thing on the board to an accent test — LOWER is betterlower is betterProvenanceindependent leaderboard74 modelsscores Aug 2, 2026
1Azure Speech 06 2026Microsoft7.52——Aug 2, 2026leaderboard2Zipformer cr CTC Transducer XL 290MSounds Good AI0.3B7.64—~42Aug 2, 2026leaderboard3MOSS Transcribe Preview 2BOpenMOSS2.4B7.72—~5.1Aug 2, 2026leaderboard4Modulate VfastModulate8.09——Aug 2, 2026leaderboard5Granite Speech 4.1 2b NARIBM2.3B8.15—~5.6Aug 2, 2026leaderboard6Hojo ASR V1Hojo AI5.2B8.15—~2.5Aug 2, 2026leaderboard7ARK ASR 3BAutoArk AI4.1B8.23—~3.2Aug 2, 2026leaderboard8Granite Speech 4.1 2bIBM2.3B8.23—~5.7Aug 2, 2026leaderboard9Higgs Audio v3 STTBoson AI2.7B8.26—~4.9Aug 2, 2026leaderboard10Granite 4.0 1b SpeechIBM2.3B8.33—~5.7Aug 2, 2026leaderboard11Higgs Audio v3 8b STT v2Boson AI8.9B8.45—~1.5Aug 2, 2026leaderboard12Resonant 1Reson88.79——Aug 2, 2026leaderboard13Scribe v2ElevenLabs8.9——Aug 2, 2026leaderboard14Resonant 1 FlashReson88.96——Aug 2, 2026leaderboard15MOSS Transcribe DiarizeOpenMOSS0.9B9.02—~16Aug 2, 2026leaderboard16Zipformer Transducer XL 290MSounds Good AI0.3B9.02—~50Aug 2, 2026leaderboard17Granite Speech 3.3 8bIBM8.6B9.07—~1.7Aug 2, 2026leaderboard18Scribe v1Zoom9.17——Aug 2, 2026leaderboard19Phi 4 Multimodal InstructMicrosoft5.6B9.35—~2.6Aug 2, 2026leaderboard20ARK ASR 0.6BAutoArk AI1.3B9.69—~12Aug 2, 2026leaderboard21Qwen3 ASR 1.7B HFQwen2B9.73—~7.5Aug 2, 2026leaderboard22Granite Speech 3.3 2bIBM3B9.73—~5.1Aug 2, 2026leaderboard23Qwen3 ASR 1.7BQwen2B9.88—~7.7Aug 2, 2026leaderboard24Universal 3 5 ProAssemblyAI10.02——Aug 2, 2026leaderboard25Canary Qwen 2.5bNVIDIA2.6B10.04—~6.1Aug 2, 2026leaderboard26Voxtral Small 24B 2507Mistral AI24.3B10.1634~0.66Aug 2, 2026leaderboard27Voxtral Mini 3B 2507Mistral AI5B10.36—~3.3Aug 2, 2026leaderboard28Cohere Transcribe 03 2026Cohere Labs2.1B10.38—~7.9Aug 2, 2026leaderboard29Solaria 3Gladia10.47——Aug 2, 2026leaderboard30STT 2.6b enKyutai2.6B10.64—~6.5Aug 2, 2026leaderboard31GLM ASR Nano 2512Z.ai2.3B10.68—~7.4Aug 2, 2026leaderboard32Lite Whisper Large v3 AccEfficient Speech1.4B10.69—~12Aug 2, 2026leaderboard33Qwen3 ASR 0.6BQwen0.8B10.72—~22Aug 2, 2026leaderboard34Parakeet TDT 0.6b v3NVIDIA0.6B10.77—~28Aug 2, 2026leaderboard35Parakeet TDT 0.6b v2NVIDIA0.6B10.78—~28Aug 2, 2026leaderboard36Qwen3 ASR 0.6B HFQwen0.8B10.78—~22Aug 2, 2026leaderboard37Distil Large v3.5Distil-Whisper0.8B10.83—~22Aug 2, 2026leaderboard38Smallest AI PulseSmallest AI10.86——Aug 2, 2026leaderboard39Universal 3 ProAssemblyAI10.89——Aug 2, 2026leaderboard40Avalon v1 enAqua Voice10.98——Aug 2, 2026leaderboard41Whisper Large v3 TurboOpenAI0.8B11.07—~22Aug 2, 2026leaderboard42Canary 1b v2NVIDIA1B11.35—~18Aug 2, 2026leaderboard43Moonshine Streaming MediumUseful Sensors0.2B11.44—~76Aug 2, 2026leaderboard44Voxtral Mini 4B Realtime 2602Mistral AI4.4B11.46—~4.1Aug 2, 2026leaderboard45Whisper Large v3OpenAI1.5B11.59—~12Aug 2, 2026leaderboard46Canary 1bNVIDIA1B11.63—~18Aug 2, 2026leaderboard47Niagara 38m Batch.enApplied Brain Research38M11.83—~590Aug 2, 2026leaderboard48Speechmatics EnhancedSpeechmatics11.86——Aug 2, 2026leaderboard49Nemotron Speech Streaming en 0.6bNVIDIA0.6B12.02—~32Aug 2, 2026leaderboard50Parakeet TDT CTC 110mNVIDIA0.1B12.02—~170Aug 2, 2026leaderboard51Canary 1b FlashNVIDIA1B12.03—~19Aug 2, 2026leaderboard52VibeVoice ASR HFMicrosoft8.3B12.15—~2.3Aug 2, 2026leaderboard53Audio8 ASR 0.1BAutoArk AI0.3B12.31—~65Aug 2, 2026leaderboard54Nyra Health CrisperWhisperNyra Health1.6B12.43—~12Aug 2, 2026leaderboard55Canary 180m FlashNVIDIA0.2B12.5—~110Aug 2, 2026leaderboard56Owsm CTC v3.2 ft 1BESPnet1B12.82—~20Aug 2, 2026leaderboard57Parakeet RNNT 1.1bNVIDIA1.1B12.94—~19Aug 2, 2026leaderboard58Parakeet CTC 1.1bNVIDIA1.1B12.98—~19Aug 2, 2026leaderboard59Moonshine Streaming SmallUseful Sensors0.1B13.09—~220Aug 2, 2026leaderboard60omniASR LLM 7B v2Meta7.8B13.15——Aug 2, 2026leaderboard61Niagara 19m Batch.enApplied Brain Research20M13.18—~1300Aug 2, 2026leaderboard62Parakeet TDT 1.1bNVIDIA1.1B13.25—~19Aug 2, 2026leaderboard63Owsm CTC v3.1 1BESPnet1.1B13.28—~19Aug 2, 2026leaderboard64Owsm CTC v4 1BESPnet1B13.37—~21Aug 2, 2026leaderboard65Parakeet CTC 0.6bNVIDIA0.6B13.47—~35Aug 2, 2026leaderboard66Parakeet RNNT 0.6bNVIDIA0.6B13.58—~36Aug 2, 2026leaderboard67ASR Conformer LargescaleasrSpeechBrain0.5B14.2—~47Aug 2, 2026leaderboard68Nemotron 3.5 ASR Streaming 0.6bNVIDIA0.6B14.95—~37Aug 2, 2026leaderboard69omniASR CTC 7B v2Meta6.5B16.02——Aug 2, 2026leaderboard70Moonshine Streaming TinyUseful Sensors30M19.52—~980Aug 2, 2026leaderboard71Moonshine TinyUseful Sensors30M19.75—~990Aug 2, 2026leaderboard72MMS 1b AllMeta1B23—~37Aug 2, 2026leaderboard73Wav2vec2 Large 960h Lv60 SelfMeta0.3B23.05—~120Aug 2, 2026leaderboard74Hubert Large Ls960 ftMeta0.3B26.87—~130Aug 2, 2026leaderboard
A tilde in Score / GB VRAM means the memory figure behind it is calculated from the parameter count rather than read from a published file. Those ratios are shown to two significant figures, and rows within a few percent of each other are not distinguishable.