Start building →OverviewMultilingualCode-switchingEmotion

Speech-to-Text · Multilingual

French

Language

Request a new language

Tell us which language you want added to this benchmark.

Benchmarks

X
Y
500ms1.00s5.0%10.0%15.0%Finalize latency lower is betterWER lower is betterAssemblyAI Universal-3.5 ProOpenAI GPT-4o TranscribeAlibaba Qwen3-ASR RealtimeElevenLabs Scribe v2 Realtime → 1.98sOpenAI GPT-4o-mini TranscribeOpenAI GPT Live TranscribeGoogle Chirp 3Soniox stt-rt-v5Deepgram Nova-3Gradium ASRCartesia Ink-WhisperSmallest AI PulseDeepgram Nova-2speko.ai/benchmarks

WER

Share of words the transcript got wrong.

AssemblyAIAssemblyAIUniversal-3.6 ProTry →3.55%
AssemblyAIAssemblyAIUniversal-3.5 ProTry →3.88%
AlibabaQwen3-ASR Realtime4.1%
NariNariQwen3-ASRTry →4.84%
ElevenLabsScribe v2 RealtimeTry →5.14%
ModulateModulateVelma 2 multilingualTry →5.86%
OpenAIGPT-4o TranscribeTry →6.1%
Sonioxstt-rt-v5Try →6.28%
MetaMetaMuse Voice Transcribe 1.0Try →6.79%
OpenAIGPT Live TranscribeTry →6.85%
GoogleChirp 37.45%
GradiumASRTry →7.89%
OpenAIGPT-4o-mini TranscribeTry →8.46%
DeepgramNova-3Try →10.49%
Smallest AISmallest AIPulseTry →11.28%
DeepgramNova-2Try →13.18%
CartesiaInk-WhisperTry →16.69%
0.0%5.0%10.0%15.0%20.0%

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-2Try →208ms · 339ms
DeepgramNova-3Try →209ms · 5.21s
CartesiaInk-WhisperTry →293ms · 404ms
AssemblyAIAssemblyAIUniversal-3.5 ProTry →469ms · 1.54s
Sonioxstt-rt-v5Try →514ms · 822ms
OpenAIGPT-4o-mini TranscribeTry →820ms · 954ms
OpenAIGPT Live TranscribeTry →836ms · 959ms
Smallest AISmallest AIPulseTry →879ms · 984ms
OpenAIGPT-4o TranscribeTry →919ms · 1.13s
AlibabaQwen3-ASR Realtime924ms · 1.12s
GoogleChirp 31.09s · 1.59s
GradiumASRTry →1.17s · 1.35s
ElevenLabsScribe v2 RealtimeTry →1.98s · 2.21s
0ms300ms600ms900ms1.20s

p50 p90

Time to first token

lower is better

How long before the first word of the transcript arrives.

AssemblyAIAssemblyAIUniversal-3.5 ProTry →1.47s · 2.77s
DeepgramNova-2Try →1.84s · 3.14s
Sonioxstt-rt-v5Try →1.91s · 3.34s
DeepgramNova-3Try →2.08s · 3.14s
OpenAIGPT Live TranscribeTry →2.17s · 3.47s
ElevenLabsScribe v2 RealtimeTry →2.21s · 4.13s
Smallest AISmallest AIPulseTry →2.40s · 3.61s
GradiumASRTry →2.43s · 4.04s
CartesiaInk-WhisperTry →4.92s · 6.56s
GoogleChirp 36.06s · 8.05s
OpenAIGPT-4o TranscribeTry →6.58s · 9.59s
OpenAIGPT-4o-mini TranscribeTry →6.68s · 10.2s
AlibabaQwen3-ASR Realtime6.88s · 8.51s
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

$1.00
$2.00
$2.10
$3.00
$4.00
$4.80
$5.40
$6.00
$6.50
$7.50
$7.50
$10.40
$16.00
$17.00
NariNariQwen3-ASRSonioxstt-rt-v5AlibabaQwen3-ASROpenAIGPT-4o-mini TranscribeSmallest AISmallest AIPulseDeepgramNova-3AlibabaQwen3-ASR RealtimeOpenAIGPT-4o TranscribeElevenLabsScribe v2 RealtimeAssemblyAIAssemblyAIUniversal-3.5 ProAssemblyAIAssemblyAIUniversal-3.6 ProGradiumASRGoogleChirp 3OpenAIGPT Live Transcribe