Start building →OverviewMultilingualCode-switchingEmotion

Speech-to-Text · Multilingual

Thai

Language

Request a new language

Tell us which language you want added to this benchmark.

Benchmarks

X
Y
50ms100ms200ms500ms1.00s2.00s4.0%6.0%8.0%10.0%Finalize latency lower is better · log scaleCER lower is betterOpenAI GPT TranscribeAlibaba Qwen3-ASR RealtimeOpenAI GPT Live TranscribeSoniox stt-rt-v5ElevenLabs Scribe v2 RealtimeModulate Velma 2 multilingualNari Qwen3-ASR FastOpenAI GPT-4o mini TranscribeOpenAI GPT-4o TranscribeGoogle Chirp 3Gladia Solaria-1Deepgram Nova-3xAI Grok STTCartesia Ink-Whisper ↓ 25.1%speko.ai/benchmarks

CER

Character error rate — word scoring mismeasures this language.

MetaMetaMuse Voice Transcribe 1.0Try →3.8%
ModulateModulateVelma 2 multilingualTry →3.9%
OpenAIGPT TranscribeTry →4.3%
AlibabaQwen3-ASR RealtimeTry →4.4%
OpenAIGPT Live TranscribeTry →4.5%
Sonioxstt-rt-v5Try →4.7%
ElevenLabsScribe v2 RealtimeTry →4.9%
NariNariQwen3-ASRTry →5.1%
NariNariQwen3-ASR FastTry →5.2%
OpenAIGPT-4o TranscribeTry →5.7%
OpenAIGPT-4o mini TranscribeTry →6.9%
DeepgramNova-3Try →8.0%
GoogleChirp 38.2%
GladiaSolaria-1Try →9.2%
xAIGrok STTTry →9.8%
CartesiaInk-WhisperTry →25.1%
0.0%10.0%20.0%30.0%

Finalize latency

Region

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

NariNariQwen3-ASR FastTry →37ms · 69ms
DeepgramNova-3Try →82ms · 167ms
CartesiaInk-WhisperTry →285ms · 658ms
OpenAIGPT TranscribeTry →798ms · 1.10s
GoogleChirp 3825ms · 2.57s
OpenAIGPT-4o mini TranscribeTry →827ms · 1.03s
AlibabaQwen3-ASR RealtimeTry →861ms · 1.08s
OpenAIGPT-4o TranscribeTry →877ms · 1.43s
GladiaSolaria-1Try →906ms · 2.43s
xAIGrok STTTry →1.17s · 1.39s
OpenAIGPT Live TranscribeTry →1.29s · 1.80s
ElevenLabsScribe v2 RealtimeTry →1.65s · 1.77s
Sonioxstt-rt-v5Try →2.11s · 2.34s
ModulateModulateVelma 2 multilingualTry →2.92s · 3.25s
0ms300ms600ms900ms1.20s

p50 p90

Time to first token

Region

How long before the first word of the transcript arrives.

DeepgramNova-3Try →1.02s · 2.00s
xAIGrok STTTry →1.03s · 2.02s
Sonioxstt-rt-v5Try →1.54s · 2.04s
ModulateModulateVelma 2 multilingualTry →1.56s · 2.07s
GladiaSolaria-1Try →1.60s · 2.99s
OpenAIGPT Live TranscribeTry →1.70s · 2.12s
NariNariQwen3-ASR FastTry →1.74s · 1.75s
ElevenLabsScribe v2 RealtimeTry →2.31s · 2.41s
GoogleChirp 34.83s · 6.48s
CartesiaInk-WhisperTry →5.52s · 9.98s
OpenAIGPT-4o TranscribeTry →6.09s · 11.1s
OpenAIGPT-4o mini TranscribeTry →6.18s · 11.2s
OpenAIGPT TranscribeTry →6.26s · 11.2s
AlibabaQwen3-ASR RealtimeTry →7.09s · 7.55s
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
$2.00
$2.10
$3.70
$4.50
$4.80
$5.40
$6.00
$6.50
$12.50
$16.00
$17.00
NariNariQwen3-ASRNariNariQwen3-ASR FastSonioxstt-rt-v5AlibabaQwen3-ASRElevenLabsScribe v2OpenAIGPT TranscribeDeepgramNova-3AlibabaQwen3-ASR RealtimeOpenAIGPT-4o TranscribeElevenLabsScribe v2 RealtimeGladiaSolaria-1GoogleChirp 3OpenAIGPT Live Transcribe