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
Open ASR RTFx
Speed as a multiple of real time on the same harness — 100 means an hour of audio in 36 seconds. Hardware-dependent, so it compares models on one rig, not machineshigher is betterProvenanceindependent leaderboard62 modelsscores Aug 1, 2026
1Parakeet TDT CTC 110mNVIDIA0.1B6118.84—~87000Aug 1, 2026leaderboard2Parakeet TDT 0.6b v3NVIDIA0.6B6098.2—~16000Aug 1, 2026leaderboard3Parakeet TDT 0.6b v2NVIDIA0.6B6038.1—~16000Aug 1, 2026leaderboard4Parakeet CTC 0.6bNVIDIA0.6B5883.91—~15000Aug 1, 2026leaderboard5Parakeet RNNT 0.6bNVIDIA0.6B5406.69—~14000Aug 1, 2026leaderboard6Parakeet CTC 1.1bNVIDIA1.1B5014.5—~7300Aug 1, 2026leaderboard7Parakeet TDT 1.1bNVIDIA1.1B4528.72—~6600Aug 1, 2026leaderboard8Moonshine Streaming TinyUseful Sensors30M4375.16—~220000Aug 1, 2026leaderboard9Parakeet RNNT 1.1bNVIDIA1.1B4121.79—~6000Aug 1, 2026leaderboard10Niagara 38m Batch.enApplied Brain Research38M4048.7—~200000Aug 1, 2026leaderboard11Niagara 19m Batch.enApplied Brain Research20M3735.48—~370000Aug 1, 2026leaderboard12Moonshine TinyUseful Sensors30M3733.06—~190000Aug 1, 2026leaderboard13Moonshine Streaming SmallUseful Sensors0.1B3205.97—~53000Aug 1, 2026leaderboard14Hubert Large Ls960 ftMeta0.3B3023.98—~15000Aug 1, 2026leaderboard15Wav2vec2 Large 960h Lv60 SelfMeta0.3B3009.97—~15000Aug 1, 2026leaderboard16Moonshine Streaming MediumUseful Sensors0.2B2681.01—~18000Aug 1, 2026leaderboard17Canary 180m FlashNVIDIA0.2B2484.22—~23000Aug 1, 2026leaderboard18Canary 1b FlashNVIDIA1B2126.07—~3400Aug 1, 2026leaderboard19Granite Speech 4.1 2b NARIBM2.3B2079.32—~1400Aug 1, 2026leaderboard20MMS 1b AllMeta1B1954.03—~3100Aug 1, 2026leaderboard21Canary 1b v2NVIDIA1B1821.44—~2900Aug 1, 2026leaderboard22Nemotron 3.5 ASR Streaming 0.6bNVIDIA0.6B1489.58—~3700Aug 1, 2026leaderboard23Nemotron Speech Streaming en 0.6bNVIDIA0.6B1071.38—~2800Aug 1, 2026leaderboard24Cohere Transcribe 03 2026Cohere Labs2.1B915.58—~690Aug 1, 2026leaderboard25Distil Large v3.5Distil-Whisper0.8B874—~1700Aug 1, 2026leaderboard26Canary Qwen 2.5bNVIDIA2.6B861.47—~530Aug 1, 2026leaderboard27Owsm CTC v3.1 1BESPnet1.1B816.21—~1200Aug 1, 2026leaderboard28Qwen3 ASR 1.7B HFQwen2B796.19—~620Aug 1, 2026leaderboard29Whisper Large v3 TurboOpenAI0.8B782.56—~1600Aug 1, 2026leaderboard30Canary 1bNVIDIA1B765.62—~1200Aug 1, 2026leaderboard31Owsm CTC v4 1BESPnet1B764.69—~1200Aug 1, 2026leaderboard32Qwen3 ASR 0.6B HFQwen0.8B730.23—~1500Aug 1, 2026leaderboard33Audio8 ASR 0.1BAutoArk AI0.3B709.29—~3700Aug 1, 2026leaderboard34Owsm CTC v3.2 ft 1BESPnet1B692.27—~1100Aug 1, 2026leaderboard35ARK ASR 0.6BAutoArk AI1.3B672.19—~820Aug 1, 2026leaderboard36Granite 4.0 1b SpeechIBM2.3B661.17—~460Aug 1, 2026leaderboard37Granite Speech 4.1 2bIBM2.3B546.75—~380Aug 1, 2026leaderboard38omniASR CTC 7B v2Meta6.5B527.58——Aug 1, 2026leaderboard39Granite Speech 3.3 2bIBM3B507.1—~270Aug 1, 2026leaderboard40ARK ASR 3BAutoArk AI4.1B484.29—~190Aug 1, 2026leaderboard41Whisper Large v3OpenAI1.5B462.22—~490Aug 1, 2026leaderboard42Qwen3 ASR 0.6BQwen0.8B438.74—~900Aug 1, 2026leaderboard43Qwen3 ASR 1.7BQwen2B394.13—~310Aug 1, 2026leaderboard44MOSS Transcribe DiarizeOpenMOSS0.9B382.37—~670Aug 1, 2026leaderboard45GLM ASR Nano 2512Z.ai2.3B332.98—~230Aug 1, 2026leaderboard46Granite Speech 3.3 8bIBM8.6B263.67—~49Aug 1, 2026leaderboard47VibeVoice ASR HFMicrosoft8.3B221.21—~42Aug 1, 2026leaderboard48Lite Whisper Large v3 AccEfficient Speech1.4B203.91—~230Aug 1, 2026leaderboard49Voxtral Mini 3B 2507Mistral AI5B179.74—~57Aug 1, 2026leaderboard50Phi 4 Multimodal InstructMicrosoft5.6B162.65—~46Aug 1, 2026leaderboard51Zipformer cr CTC Transducer XL 290MSounds Good AI0.3B159.77—~890Aug 1, 2026leaderboard52MOSS Transcribe Preview 2BOpenMOSS2.4B148.51—~98Aug 1, 2026leaderboard53omniASR LLM 7B v2Meta7.8B142.87——Aug 1, 2026leaderboard54Zipformer Transducer XL 290MSounds Good AI0.3B141.22—~780Aug 1, 2026leaderboard55Higgs Audio v3 8b STT v2Boson AI8.9B139.1—~25Aug 1, 2026leaderboard56STT 2.6b enKyutai2.6B133.17—~81Aug 1, 2026leaderboard57Higgs Audio v3 STTBoson AI2.7B110.12—~65Aug 1, 2026leaderboard58Voxtral Mini 4B Realtime 2602Mistral AI4.4B105.14—~38Aug 1, 2026leaderboard59Voxtral Small 24B 2507Mistral AI24.3B100.08334~6.5Aug 1, 2026leaderboard60Hojo ASR V1Hojo AI5.2B73.8—~23Aug 1, 2026leaderboard61ASR Conformer LargescaleasrSpeechBrain0.5B72.36—~240Aug 1, 2026leaderboard62Nyra Health CrisperWhisperNyra Health1.6B33.25—~33Aug 1, 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.