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

Clean read speech

Word error rate on LibriSpeech test-clean — audiobooks read aloud in good conditions. The easy case, and the floor every model should clear — LOWER is betterlower is better
Provenanceindependent leaderboard74 modelsscores Aug 2, 2026
1Modulate VfastModulate0.92Aug 2, 2026leaderboard2Resonant 1Reson80.92Aug 2, 2026leaderboard3Resonant 1 FlashReson80.94Aug 2, 2026leaderboard4Higgs Audio v3 8b STT v2Boson AI8.9B0.95~0.17Aug 2, 2026leaderboard5Cohere Transcribe 03 2026Cohere Labs2.1B0.96~0.73Aug 2, 2026leaderboard6Granite Speech 4.1 2bIBM2.3B1.02~0.7Aug 2, 2026leaderboard7ARK ASR 3BAutoArk AI4.1B1.03~0.4Aug 2, 2026leaderboard8Granite Speech 4.1 2b NARIBM2.3B1.04~0.72Aug 2, 2026leaderboard9Scribe v2ElevenLabs1.05Aug 2, 2026leaderboard10Parakeet TDT 1.1bNVIDIA1.1B1.05~1.5Aug 2, 2026leaderboard11Higgs Audio v3 STTBoson AI2.7B1.07~0.63Aug 2, 2026leaderboard12Granite 4.0 1b SpeechIBM2.3B1.1~0.76Aug 2, 2026leaderboard13Universal 3 5 ProAssemblyAI1.1Aug 2, 2026leaderboard14Universal 3 ProAssemblyAI1.1Aug 2, 2026leaderboard15Granite Speech 3.3 8bIBM8.6B1.11~0.2Aug 2, 2026leaderboard16Scribe v1Zoom1.12Aug 2, 2026leaderboard17Parakeet RNNT 1.1bNVIDIA1.1B1.15~1.7Aug 2, 2026leaderboard18Canary 1bNVIDIA1B1.15~1.8Aug 2, 2026leaderboard19MOSS Transcribe Preview 2BOpenMOSS2.4B1.15~0.76Aug 2, 2026leaderboard20Azure Speech 06 2026Microsoft1.19Aug 2, 2026leaderboard21Canary 1b FlashNVIDIA1B1.2~1.9Aug 2, 2026leaderboard22Granite Speech 3.3 2bIBM3B1.21~0.64Aug 2, 2026leaderboard23Canary Qwen 2.5bNVIDIA2.6B1.23~0.75Aug 2, 2026leaderboard24Voxtral Small 24B 2507Mistral AI24.3B1.234~0.08Aug 2, 2026leaderboard25Qwen3 ASR 1.7BQwen2B1.24~0.96Aug 2, 2026leaderboard26Parakeet RNNT 0.6bNVIDIA0.6B1.25~3.3Aug 2, 2026leaderboard27Qwen3 ASR 1.7B HFQwen2B1.26~0.98Aug 2, 2026leaderboard28Avalon v1 enAqua Voice1.27Aug 2, 2026leaderboard29Parakeet TDT 0.6b v2NVIDIA0.6B1.27~3.3Aug 2, 2026leaderboard30Hojo ASR V1Hojo AI5.2B1.3~0.4Aug 2, 2026leaderboard31Smallest AI PulseSmallest AI1.3Aug 2, 2026leaderboard32Zipformer cr CTC Transducer XL 290MSounds Good AI0.3B1.31~7.3Aug 2, 2026leaderboard33Solaria 3Gladia1.31Aug 2, 2026leaderboard34omniASR LLM 7B v2Meta7.8B1.35Aug 2, 2026leaderboard35Phi 4 Multimodal InstructMicrosoft5.6B1.35~0.38Aug 2, 2026leaderboard36Wav2vec2 Large 960h Lv60 SelfMeta0.3B1.36~6.8Aug 2, 2026leaderboard37STT 2.6b enKyutai2.6B1.37~0.84Aug 2, 2026leaderboard38Voxtral Mini 3B 2507Mistral AI5B1.47~0.47Aug 2, 2026leaderboard39ARK ASR 0.6BAutoArk AI1.3B1.47~1.8Aug 2, 2026leaderboard40Parakeet TDT 0.6b v3NVIDIA0.6B1.51~4Aug 2, 2026leaderboard41Canary 180m FlashNVIDIA0.2B1.52~14Aug 2, 2026leaderboard42Parakeet CTC 1.1bNVIDIA1.1B1.52~2.2Aug 2, 2026leaderboard43VibeVoice ASR HFMicrosoft8.3B1.55~0.3Aug 2, 2026leaderboard44Hubert Large Ls960 ftMeta0.3B1.55~7.8Aug 2, 2026leaderboard45Whisper Large v3OpenAI1.5B1.55~1.6Aug 2, 2026leaderboard46Parakeet CTC 0.6bNVIDIA0.6B1.57~4.1Aug 2, 2026leaderboard47Voxtral Mini 4B Realtime 2602Mistral AI4.4B1.61~0.58Aug 2, 2026leaderboard48Lite Whisper Large v3 AccEfficient Speech1.4B1.63~1.9Aug 2, 2026leaderboard49MOSS Transcribe DiarizeOpenMOSS0.9B1.66~2.9Aug 2, 2026leaderboard50Zipformer Transducer XL 290MSounds Good AI0.3B1.67~9.3Aug 2, 2026leaderboard51Moonshine Streaming MediumUseful Sensors0.2B1.67~11Aug 2, 2026leaderboard52GLM ASR Nano 2512Z.ai2.3B1.69~1.2Aug 2, 2026leaderboard53Qwen3 ASR 0.6BQwen0.8B1.69~3.4Aug 2, 2026leaderboard54Qwen3 ASR 0.6B HFQwen0.8B1.7~3.4Aug 2, 2026leaderboard55Speechmatics EnhancedSpeechmatics1.73Aug 2, 2026leaderboard56Canary 1b v2NVIDIA1B1.76~2.8Aug 2, 2026leaderboard57ASR Conformer LargescaleasrSpeechBrain0.5B1.83~6.1Aug 2, 2026leaderboard58Nemotron Speech Streaming en 0.6bNVIDIA0.6B1.88~4.9Aug 2, 2026leaderboard59Owsm CTC v3.1 1BESPnet1.1B1.9~2.8Aug 2, 2026leaderboard60Distil Large v3.5Distil-Whisper0.8B1.93~3.9Aug 2, 2026leaderboard61Nyra Health CrisperWhisperNyra Health1.6B1.99~2Aug 2, 2026leaderboard62Parakeet TDT CTC 110mNVIDIA0.1B2.01~29Aug 2, 2026leaderboard63Moonshine Streaming SmallUseful Sensors0.1B2.07~35Aug 2, 2026leaderboard64Owsm CTC v4 1BESPnet1B2.1~3.3Aug 2, 2026leaderboard65omniASR CTC 7B v2Meta6.5B2.12Aug 2, 2026leaderboard66Whisper Large v3 TurboOpenAI0.8B2.13~4.3Aug 2, 2026leaderboard67Owsm CTC v3.2 ft 1BESPnet1B2.17~3.4Aug 2, 2026leaderboard68Audio8 ASR 0.1BAutoArk AI0.3B2.7~14Aug 2, 2026leaderboard69Nemotron 3.5 ASR Streaming 0.6bNVIDIA0.6B2.83~7.1Aug 2, 2026leaderboard70MMS 1b AllMeta1B3.14~5Aug 2, 2026leaderboard71Niagara 38m Batch.enApplied Brain Research38M3.75~190Aug 2, 2026leaderboard72Moonshine Streaming TinyUseful Sensors30M3.98~200Aug 2, 2026leaderboard73Moonshine TinyUseful Sensors30M4.08~200Aug 2, 2026leaderboard74Niagara 19m Batch.enApplied Brain Research20M4.45~450Aug 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.