Speech-to-Text · Multilingual

Japanese

Language

Request a new language

Tell us which language you want added to this benchmark.

Benchmarks

X
Y

CER

Character error rate — word scoring mismeasures this language.

Sonioxstt-rt-v5Try →1.8%
AssemblyAIAssemblyAIUniversal-3.5 Pro3.4%
OpenAIGPT TranscribeTry →3.7%
AlibabaQwen3-ASRTry →3.7%
ElevenLabsScribe v2 RealtimeTry →4.4%
OpenAIGPT-4o mini TranscribeTry →4.5%
ModulateModulateVelma 2 multilingual4.8%
OpenAIGPT-4o TranscribeTry →5.2%
Smallest AISmallest AIPulse6.1%
OpenAIGPT Live Transcribe6.7%
GladiaSolaria-1Try →17.6%
DeepgramNova-3Try →22.0%
xAIGrok STTTry →34.6%
CartesiaInk-WhisperTry →35.0%
0.0%20.0%40.0%60.0%

Finalize latency

lower is better

Time from the caller stopping to the finalized transcript, not the first partial — the agent cannot reply until it lands.

DeepgramNova-3Try →314ms · 1.69s
CartesiaInk-WhisperTry →368ms · 542ms
Sonioxstt-rt-v5Try →483ms · 1.15s
AssemblyAIAssemblyAIUniversal-3.5 Pro568ms · 1.06s
OpenAIGPT-4o mini TranscribeTry →821ms · 996ms
Smallest AISmallest AIPulse824ms · 8.25s
OpenAIGPT TranscribeTry →873ms · 1.11s
OpenAIGPT-4o TranscribeTry →901ms · 1.24s
AlibabaQwen3-ASRTry →964ms · 1.05s
GladiaSolaria-1Try →1.04s · 2.61s
xAIGrok STTTry →1.29s · 2.35s
ElevenLabsScribe v2 RealtimeTry →2.02s · 2.22s
0ms300ms600ms900ms1.20s

p50 p90

Time to first token

lower is better

How long before the first word of the transcript arrives.

DeepgramNova-3Try →2.20s · 4.42s
AssemblyAIAssemblyAIUniversal-3.5 Pro2.36s · 4.46s
Sonioxstt-rt-v5Try →2.54s · 4.21s
xAIGrok STTTry →2.63s · 5.71s
GladiaSolaria-1Try →2.74s · 5.38s
ElevenLabsScribe v2 RealtimeTry →3.10s · 4.20s
Smallest AISmallest AIPulse3.57s · 4.58s
OpenAIGPT-4o mini TranscribeTry →5.69s · 10.8s
AlibabaQwen3-ASRTry →5.75s · 9.80s
OpenAIGPT-4o TranscribeTry →5.84s · 10.6s
OpenAIGPT TranscribeTry →5.89s · 10.9s
CartesiaInk-WhisperTry →5.95s · 11.6s
0ms2.00s4.00s6.00s8.00s

p50 p90

Cost

lower is better

$ per 1,000 minutes of audio · vendor list price, the same in every language