STT

#	Model	Overall	WER · batch / stream	End-of-turn · faster ▸	Cost / min
1	PulseSmallest AI⚠	90	5.1% / 8.0%	183ms †	~$0.005
2	GPT-4o TranscribeOpenAI⚠	87	2.3% / 5.8%	409ms	$0.006
3	stt-rt-v5Soniox⚠	82	7.5% / 7.3%	269ms	$0.002
4	Grok STTxAI⚠	80	4.8% / 10.9%	452ms	—
5	GPT-4o-mini TranscribeOpenAI⚠	75	2.7% / 6.4%	679ms	$0.003
6	Qwen3-ASRAlibaba⚠	74	2.8% / 4.0%	710ms	—
7	Nova-3Deepgram	74	9.8% / 12.9%	368ms	$0.0048
8	Gradium ASRGradium⚠	73	13.7% / 17.6%	195ms	—
9	Ink-2Cartesia⚠	63	11.0% / 9.9%	575ms	$0.0022
10	Chirp 3Google	60	3.9% / 7.4%	995ms	—
11	Scribe v2ElevenLabs⚠	54	3.1% / 3.4%	1750ms	$0.0067

WER · batch / streaminglower is better · faint = live stream

GPT-4o Transcribe2.3% / 5.8%

GPT-4o-mini Transcribe2.7% / 6.4%

Qwen3-ASR2.8% / 4.0%

Scribe v23.1% / 3.4%

Chirp 33.9% / 7.4%

Grok STT4.8% / 10.9%

Pulse5.1% / 8.0%

stt-rt-v57.5% / 7.3%

Nova-39.8% / 12.9%

Ink-211.0% / 9.9%

Gradium ASR13.7% / 17.6%

End-of-turn latencyfaster is better

Pulse183ms †

Gradium ASR195ms

stt-rt-v5269ms

Nova-3368ms

GPT-4o Transcribe409ms

Grok STT452ms

Ink-2575ms

GPT-4o-mini Transcribe679ms

Qwen3-ASR710ms

Chirp 3995ms

Scribe v21750ms

Cost / mincheaper is better

stt-rt-v5$0.002

Ink-2$0.0022

GPT-4o-mini Transcribe$0.003

Nova-3$0.0048

Pulse~$0.005

GPT-4o Transcribe$0.006

Scribe v2$0.0067