Start building →OverviewMultilingualCode-switchingEmotion

Speech-to-Text · Multilingual

Korean

Language

Request a new language

Tell us which language you want added to this benchmark.

Benchmarks

X
Y
2ms5ms10ms20ms50ms100ms200ms500ms1.00s2.00s5.0%6.0%Finalize latency lower is better · log scaleCER lower is betterSoniox stt-rt-v5ElevenLabs Scribe v2 RealtimexAI Grok STTAlibaba Qwen3-ASROpenAI GPT-4o TranscribeGladia Solaria-1OpenAI GPT TranscribeDeepgram Nova-3 ↓ 11.6%speko.ai/benchmarks

CER

Character error rate — word scoring mismeasures this language.

Sonioxstt-rt-v5Try →4.2%
ElevenLabsScribe v2 RealtimeTry →4.5%
xAIGrok STT4.7%
AlibabaQwen3-ASR4.7%
OpenAIGPT-4o TranscribeTry →4.9%
GladiaSolaria-1Try →5.9%
OpenAIGPT TranscribeTry →6.4%
DeepgramNova-3Try →11.6%
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.

Sonioxstt-rt-v5Try →2ms · 1.12s
DeepgramNova-3Try →177ms · 248ms
GladiaSolaria-1Try →608ms · 2.33s
AlibabaQwen3-ASR804ms · 900ms
OpenAIGPT TranscribeTry →820ms · 1.07s
OpenAIGPT-4o TranscribeTry →844ms · 1.10s
xAIGrok STT1.08s · 1.19s
ElevenLabsScribe v2 RealtimeTry →1.62s · 1.74s
0ms300ms600ms900ms1.20s

p50 p90

Time to first token

lower is better

How long before the first word of the transcript arrives.

Sonioxstt-rt-v5Try →2.36s · 3.55s
xAIGrok STT2.54s · 4.06s
DeepgramNova-3Try →2.56s · 3.96s
GladiaSolaria-1Try →2.70s · 3.91s
ElevenLabsScribe v2 RealtimeTry →2.92s · 4.21s
OpenAIGPT-4o TranscribeTry →6.71s · 10.7s
OpenAIGPT TranscribeTry →6.75s · 10.5s
AlibabaQwen3-ASR7.39s · 9.14s
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

$2.00
$2.10
$4.50
$4.80
$6.00
$6.50
$12.50
Sonioxstt-rt-v5AlibabaQwen3-ASROpenAIGPT TranscribeDeepgramNova-3OpenAIGPT-4o TranscribeElevenLabsScribe v2 RealtimeGladiaSolaria-1