Leaderboards

Benchmarks

Every score carries its origin — leaderboard is independent,vendor is self-reported. History is kept, never overwritten.

Reasoning280Coding266Math174Agentic110Chat515Speech654
Accented speech74Clean read speech74European-accented speech74Financial calls74Harder read speech74Open ASR RTFx62Open ASR WER74Podcasts and video74Recorded meetings74

European-accented speech

Word error rate on VoxPopuli — European Parliament proceedings in English, mostly from non-native speakers — LOWER is betterlower is better
Provenanceindependent leaderboard74 modelsscores Aug 2, 2026
1Scribe v2ElevenLabs1.41Aug 2, 2026leaderboard2Universal 3 ProAssemblyAI1.83Aug 2, 2026leaderboard3Avalon v1 enAqua Voice1.86Aug 2, 2026leaderboard4Universal 3 5 ProAssemblyAI1.92Aug 2, 2026leaderboard5Voxtral Small 24B 2507Mistral AI24.3B27~0.13Aug 2, 2026leaderboard6VibeVoice ASR HFMicrosoft8.3B2.02~0.39Aug 2, 2026leaderboard7GLM ASR Nano 2512Z.ai2.3B2.05~1.4Aug 2, 2026leaderboard8Resonant 1 FlashReson82.19Aug 2, 2026leaderboard9Resonant 1Reson82.19Aug 2, 2026leaderboard10Voxtral Mini 3B 2507Mistral AI5B2.22~0.7Aug 2, 2026leaderboard11Nemotron Speech Streaming en 0.6bNVIDIA0.6B2.4~6.3Aug 2, 2026leaderboard12Distil Large v3.5Distil-Whisper0.8B2.52~5Aug 2, 2026leaderboard13Voxtral Mini 4B Realtime 2602Mistral AI4.4B2.6~0.94Aug 2, 2026leaderboard14Solaria 3Gladia2.76Aug 2, 2026leaderboard15ARK ASR 0.6BAutoArk AI1.3B2.76~3.4Aug 2, 2026leaderboard16Speechmatics EnhancedSpeechmatics2.83Aug 2, 2026leaderboard17Qwen3 ASR 1.7B HFQwen2B2.85~2.2Aug 2, 2026leaderboard18Qwen3 ASR 0.6B HFQwen0.8B2.9~5.8Aug 2, 2026leaderboard19Higgs Audio v3 8b STT v2Boson AI8.9B2.94~0.52Aug 2, 2026leaderboard20Qwen3 ASR 1.7BQwen2B3.01~2.3Aug 2, 2026leaderboard21Qwen3 ASR 0.6BQwen0.8B3.07~6.3Aug 2, 2026leaderboard22Hojo ASR V1Hojo AI5.2B3.11~0.95Aug 2, 2026leaderboard23Parakeet TDT 0.6b v3NVIDIA0.6B3.19~8.4Aug 2, 2026leaderboard24Parakeet CTC 0.6bNVIDIA0.6B3.19~8.4Aug 2, 2026leaderboard25Parakeet CTC 1.1bNVIDIA1.1B3.24~4.7Aug 2, 2026leaderboard26Lite Whisper Large v3 AccEfficient Speech1.4B3.35~3.8Aug 2, 2026leaderboard27Moonshine Streaming MediumUseful Sensors0.2B3.37~22Aug 2, 2026leaderboard28Parakeet RNNT 0.6bNVIDIA0.6B3.41~9Aug 2, 2026leaderboard29STT 2.6b enKyutai2.6B3.53~2.2Aug 2, 2026leaderboard30ARK ASR 3BAutoArk AI4.1B3.55~1.4Aug 2, 2026leaderboard31Canary 180m FlashNVIDIA0.2B3.57~32Aug 2, 2026leaderboard32MOSS Transcribe DiarizeOpenMOSS0.9B3.6~6.3Aug 2, 2026leaderboard33Parakeet TDT 1.1bNVIDIA1.1B3.62~5.2Aug 2, 2026leaderboard34Parakeet RNNT 1.1bNVIDIA1.1B3.64~5.3Aug 2, 2026leaderboard35Parakeet TDT CTC 110mNVIDIA0.1B3.72~53Aug 2, 2026leaderboard36Higgs Audio v3 STTBoson AI2.7B3.73~2.2Aug 2, 2026leaderboard37Modulate VfastModulate3.74Aug 2, 2026leaderboard38Moonshine Streaming SmallUseful Sensors0.1B3.76~63Aug 2, 2026leaderboard39Parakeet TDT 0.6b v2NVIDIA0.6B3.78~9.9Aug 2, 2026leaderboard40Smallest AI PulseSmallest AI3.85Aug 2, 2026leaderboard41Azure Speech 06 2026Microsoft3.87Aug 2, 2026leaderboard42Scribe v1Zoom3.95Aug 2, 2026leaderboard43Canary 1bNVIDIA1B4.01~6.4Aug 2, 2026leaderboard44omniASR LLM 7B v2Meta7.8B4.03Aug 2, 2026leaderboard45ASR Conformer LargescaleasrSpeechBrain0.5B4.06~14Aug 2, 2026leaderboard46Zipformer Transducer XL 290MSounds Good AI0.3B4.08~23Aug 2, 2026leaderboard47Owsm CTC v4 1BESPnet1B4.09~6.5Aug 2, 2026leaderboard48Canary Qwen 2.5bNVIDIA2.6B4.14~2.5Aug 2, 2026leaderboard49Canary 1b FlashNVIDIA1B4.17~6.6Aug 2, 2026leaderboard50Phi 4 Multimodal InstructMicrosoft5.6B4.18~1.2Aug 2, 2026leaderboard51Granite Speech 4.1 2bIBM2.3B4.18~2.9Aug 2, 2026leaderboard52Nemotron 3.5 ASR Streaming 0.6bNVIDIA0.6B4.24~11Aug 2, 2026leaderboard53Granite Speech 4.1 2b NARIBM2.3B4.25~2.9Aug 2, 2026leaderboard54Nyra Health CrisperWhisperNyra Health1.6B4.27~4.2Aug 2, 2026leaderboard55Zipformer cr CTC Transducer XL 290MSounds Good AI0.3B4.29~24Aug 2, 2026leaderboard56Audio8 ASR 0.1BAutoArk AI0.3B4.39~23Aug 2, 2026leaderboard57Canary 1b v2NVIDIA1B4.39~7Aug 2, 2026leaderboard58Granite 4.0 1b SpeechIBM2.3B4.41~3Aug 2, 2026leaderboard59omniASR CTC 7B v2Meta6.5B4.43Aug 2, 2026leaderboard60Owsm CTC v3.2 ft 1BESPnet1B4.45~7.1Aug 2, 2026leaderboard61Granite Speech 3.3 2bIBM3B4.52~2.4Aug 2, 2026leaderboard62Whisper Large v3OpenAI1.5B4.53~4.8Aug 2, 2026leaderboard63Granite Speech 3.3 8bIBM8.6B4.54~0.84Aug 2, 2026leaderboard64Owsm CTC v3.1 1BESPnet1.1B4.58~6.6Aug 2, 2026leaderboard65MOSS Transcribe Preview 2BOpenMOSS2.4B4.88~3.2Aug 2, 2026leaderboard66MMS 1b AllMeta1B5.1~8.1Aug 2, 2026leaderboard67Niagara 38m Batch.enApplied Brain Research38M5.16~260Aug 2, 2026leaderboard68Cohere Transcribe 03 2026Cohere Labs2.1B5.38~4.1Aug 2, 2026leaderboard69Niagara 19m Batch.enApplied Brain Research20M5.93~590Aug 2, 2026leaderboard70Wav2vec2 Large 960h Lv60 SelfMeta0.3B6.27~31Aug 2, 2026leaderboard71Moonshine TinyUseful Sensors30M6.52~330Aug 2, 2026leaderboard72Whisper Large v3 TurboOpenAI0.8B7.02~14Aug 2, 2026leaderboard73Hubert Large Ls960 ftMeta0.3B7.78~39Aug 2, 2026leaderboard74Moonshine Streaming TinyUseful Sensors30M8.11~410Aug 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.