Skip to content

Models

Speech-to-text

#	Model	Id	WER / batch	WER / stream	Finalize p50
1	Universal-3.5 Pro		2.0%	2.0%	66ms
2	stt-rt-v5		7.5%	7.3%	78ms
3	Ink-2		11.0%	9.9%	102ms
4	Nova-3		9.8%	12.9%	106ms
5	Realtime STT-1	not routable	3.3%	3.6%	139ms
6	Pulse		5.1%	8.0%	178ms
7	Scribe v2		3.1%	3.4%	233ms
8	Grok STT		4.8%	10.9%	305ms
9	Gradium ASR		8.4%	11.7%	334ms
10	Qwen3-ASR		2.8%	4.0%	424ms
11	GPT-4o-mini Transcribe		2.7%	6.4%	460ms
12	GPT-4o Transcribe		2.3%	5.8%	572ms
13	Chirp 3		3.9%	7.4%	581ms

LLM

#	Model	Id	Score	Task done	TTFT p50
1	gemma-4-31b		92	89%	192ms
2	Claude Haiku 4.5		91	100%	532ms
3	gpt-5.6-luna		88	100%	449ms
4	gpt-4.1		84	94%	640ms
5	gemini-3.5-flash-lite		83	86%	373ms
6	gpt-4.1-mini		80	100%	708ms
7	gpt-oss-120b		80	94%	195ms
8	gpt-5		78	89%	635ms
9	gpt-5-nano		74	56%	525ms
10	Claude Sonnet 5		68	89%	1.21s
11	kimi-k2.5		64	89%	1.39s
12	gpt-5-mini		64	61%	746ms
13	Llama-3.3-70B		62	56%	698ms
14	qwen-turbo		49	33%	483ms
15	Grok-4.3		43	100%	2.15s
16	DeepSeek-V4-Pro		not measured	100%	not measured

Text-to-speech

#	Model	Id	Naturalness / internal MOS	Synth p50
1	eleven_v3		3.51	475ms
2	gemini-3.1-flash-tts-preview		3.51	905ms
3	aura-2		3.47	122ms
4	sonic-3.5		3.42	126ms
5	simba-3.2		3.41	491ms
6	tts-rt-v1		3.39	398ms
7	inworld-tts-2		3.34	120ms
8	lightning_v3.1		3.25	175ms
9	grok-tts		2.96	300ms
10	default		2.69	373ms
11	speech-2.8-hd		2.60	320ms
12	arcanav3		2.59	312ms
13	gpt-4o-mini-tts		2.56	658ms
14	octave-2		2.30	441ms
15	qwen3-tts-flash		1.98	488ms

Speech-to-speech

#	Model	Id	Overall	Action	Latency p50
1	gemini-3.1-flash-live		0.77	0.78	1108ms
2	gpt-realtime		0.76	0.83	494ms
3	gpt-realtime-2.1-mini		0.75	0.83	1011ms
4	grok-voice-fast		0.72	0.44	1111ms
5	gpt-realtime-2		0.68	0.87	1103ms
6	gpt-realtime-2.1		0.63	0.56	1104ms
7	gpt-realtime-mini		0.62	0.70	614ms
8	grok-voice-think-fast		0.56	0.22	1007ms
9	gpt-4.1-nano (cascade)		0.45	1.00	not measured

Spanish Speech-to-text

#	Model	Id	WER / batch
1	GPT-4o Transcribe		3.8%
2	Scribe v2		4.2%
3	Qwen3-ASR		4.4%
4	GPT-4o-mini Transcribe		4.6%
5	Whisper-1		4.6%
6	Scribe v1		5.1%
7	Chirp 3		5.4%
8	Ink-Whisper		7.3%
9	stt-rt-v5		7.3%
10	Gradium ASR		8.2%
11	Nova-3		8.4%
12	Pulse Pro		8.4%
13	Nova-2		11.3%

Spanish Text-to-speech

#	Model	Id	Naturalness / internal MOS
1	eleven_v3		4.68
2	sonic-3.5		4.12
3	grok-tts		3.64
4	inworld-tts-2		3.48
5	octave-2		3.33
6	tts-rt-v1		3.12
7	lightning_v3.1		2.93
8	speech-2.8-hd		2.80
9	qwen3-tts-flash		2.69
10	gpt-4o-mini-tts		2.55
11	arcanav3		1.63
12	aura-2		1.34

German Speech-to-text

#	Model	Id	WER / batch
1	GPT-4o Transcribe		2.1%
2	Qwen3-ASR		3.0%
3	Whisper-1		3.2%
4	Scribe v2		4.0%
5	GPT-4o-mini Transcribe		4.1%
6	Chirp 3		5.0%
7	Gradium ASR		5.8%
8	Scribe v1		6.4%
9	Ink-Whisper		7.4%
10	Pulse Pro		9.2%
11	Nova-3		11.2%
12	Nova-2		12.0%
13	stt-rt-v5		13.4%

German Text-to-speech

#	Model	Id	Naturalness / internal MOS
1	eleven_v3		4.25
2	grok-tts		4.12
3	tts-rt-v1		3.94
4	octave-2		3.81
5	sonic-3.5		3.72
6	inworld-tts-2		3.31
7	qwen3-tts-flash		3.29
8	speech-2.8-hd		3.08
9	gpt-4o-mini-tts		2.93
10	arcanav3		1.94
11	lightning_v3.1		1.93
12	default		1.90
13	aura-2		1.39

French Speech-to-text

#	Model	Id	WER / batch
1	GPT-4o Transcribe		4.0%
2	Qwen3-ASR		4.4%
3	Whisper-1		5.8%
4	Scribe v2		6.7%
5	GPT-4o-mini Transcribe		6.8%
6	Scribe v1		7.0%
7	stt-rt-v5		8.6%
8	Nova-3		9.9%
9	Ink-Whisper		10.4%
10	Gradium ASR		11.3%
11	Pulse Pro		11.8%
12	Nova-2		12.9%

French Text-to-speech

#	Model	Id	Naturalness / internal MOS
1	eleven_v3		4.23
2	grok-tts		3.90
3	inworld-tts-2		3.86
4	octave-2		3.80
5	tts-rt-v1		3.60
6	sonic-3.5		3.57
7	gpt-4o-mini-tts		3.15
8	qwen3-tts-flash		3.11
9	speech-2.8-hd		2.57
10	lightning_v3.1		2.56
11	arcanav3		2.11
12	default		1.80
13	aura-2		1.45

Arabic Speech-to-text

#	Model	Id	CER / batch
1	GPT-4o Transcribe		2.7%
2	Whisper-1		2.8%
3	Qwen3-ASR		3.5%
4	GPT-4o-mini Transcribe		4.7%
5	stt-rt-v5		5.1%
6	Nova-3		7.1%
7	Scribe v2		7.5%
8	Scribe v1		8.4%
9	Ink-Whisper		10.2%

Arabic Text-to-speech

#	Model	Id	Naturalness / internal MOS
1	grok-tts		4.15
2	sonic-3.5		3.73
3	eleven_v3		3.33
4	inworld-tts-2		2.76

Norwegian Speech-to-text

#	Model	Id	WER / stream
1	stt-rt-v5		8.0%
2	Scribe v2		8.3%
3	Qwen3-ASR		14.5%
4	Nova-3		14.9%
5	Nova-2		18.6%

Norwegian Text-to-speech

#	Model	Id	Naturalness / internal MOS
1	sonic-3.5		4.49
2	gemini-3.1-flash-tts-preview		4.01
3	eleven_v3		3.67
4	Azure AI Speech (native)	not routable	2.87
5	gpt-4o-mini-tts		2.62
6	speech-2.8-hd		2.45

Filipino Text-to-speech

#	Model	Id	Naturalness / internal MOS
1	gemini-3.1-flash-tts-preview		4.71
2	eleven_v3		3.77
3	gpt-4o-mini-tts		3.70
4	grok-tts		3.62
5	speech-2.8-hd		3.38
6	sonic-3.5		3.37
7	inworld-tts-2		2.02

Recommended stacks

Use case	Speech-to-text	LLM	Text-to-speech
Real-time phone agent	Pulse	gemma-4-31b	sonic-3.5
Accuracy-criticalNatural conversationTool-heavy agent	Scribe v2	Claude Haiku 4.5	sonic-3.5

One key, 39 routable models, one request shape.

Get API key Explore models