Speech-to-Text Benchmark
Benchmarks
XYWER
0.0%10.0%20.0%30.0%40.0%
End of turn
lower is betterTime from the last spoken word to the final transcript, decided by the model itself.
0ms500ms1.00s1.50s2.00s2.50s
p50 p90
Time to first token
lower is betterTime from the caller starting to speak to the first word of the live transcript.
0ms1.00s2.00s3.00s4.00s5.00s
p50 p90
Live words
lower is betterTime from a word being spoken to it appearing in the live transcript for good.
0ms1.00s2.00s3.00s4.00s5.00s
p50 p90
Rewrites
lower is betterWords the live transcript takes back, per word in the final transcript.
0%40%80%120%160%
Cost
lower is betterPublished list price per thousand minutes; batch-only models show their batch rate.
ModulateVelma 2Hoppernemotron-asrNariQwen3-ASRFonadaLabsKapture v3Sonioxstt-async-v6AzureMAI-Transcribe-2HopperNemotron 3.5 TranscribeNariQwen3-ASR FastSonioxstt-rt-v5Sonioxstt-rt-v6AlibabaQwen3-ASR FlashInworldRealtime STT-1AssemblyAIUniversal-Streaming EnglishAssemblyAIUniversal-Streaming MultilingualSpeechmaticsLinden 1OpenAIGPT-4o mini TranscribeMetaMuse Voice TranscribexAIGrok Voice Transcribe 1.0xAIGrok Voice Transcribe 2.0ElevenLabsScribe v3MistralVoxtral Mini Transcribe 3ElevenLabsScribe v3 LiveElevenLabsScribe v2Omimedical-1Smallest AIPulseSpeechmaticsStandardMistralVoxtral Small Transcribe 3OpenAIGPT TranscribeOpenAIGPT-4o TranscribeFish AudioTranscribe 1 ProFish Audiotranscribe-stream-1ElevenLabsScribe v2 RealtimeSpeechmaticsEnhancedAssemblyAI3.7 Realtime AgentAssemblyAIUniversal-3.5 ProAssemblyAIUniversal-3.6 ProDeepgramFluxDeepgramNova-3DeepgramFlux MultilingualMistralVoxtral Mini Transcribe Realtime 3CartesiaInk-2GradiumASRGladiaSolaria-1OpenAIGPT Live TranscribeOpenAIGPT Realtime Whisper
