Speko
Get API key

Voice AI models and benchmarks

97 supported models. Missing benchmark results are marked “not measured”. Prices are the benchmark publisher’s vendor rates. Model availability is based on a saved catalog; live availability could not be verified.

Filter catalog

Provider

97 models

Speech-to-text32

ModelProviderId
Pulse
Smallest AI logoSmallest AI
14.5%5.1%5.9%839ms$0.0040
Nova-3
Deepgram logoDeepgram
17.0%12.0%10.8%189ms$0.0048
Flux
Deepgram logoDeepgram
19.7%not measured7.2%712ms$0.0065
flux-general-multi
Deepgram logoDeepgram
21.4%not measured8.7%854ms$0.0078
Ink-2
Cartesia logoCartesia
16.7%11.0%10.1%587ms$0.0090
Qwen3-ASR Realtime
Alibaba logoAlibaba
12.2%not measured4.0%889ms$0.0054
Chirp 3Batch only
Google logoGoogle
18.4%3.9%5.9%not measured$0.0160
Scribe v2 Realtime
ElevenLabs logoElevenLabs
12.5%not measured4.3%1.75s$0.0065
transcribe-1-proBatch only
Fish Audio logoFish Audio
not measured3.4%not measurednot measured$0.0060
transcribe-1Batch only
Fish Audio logoFish Audio
not measured3.2%not measurednot measured$0.0060
MAI-Transcribe-2Batch only
Azure logoAzure
not measured2.3%not measurednot measured$0.0017
Universal-3.5 Pro
AssemblyAI logoAssemblyAI
14.5%2.0%3.7%408ms$0.0075
Universal-3.6 Pro
AssemblyAI logoAssemblyAI
13.7%not measured3.1%505ms$0.0075
Velma 2
Modulate logoModulate
14.4%4.4%5.3%1.21s$0.0008
stt-rt-v5
Soniox logoSoniox
16.6%7.5%6.3%340ms$0.0020
GPT Transcribe
OpenAI logoOpenAI
16.8%2.5%5.8%817ms$0.0045
GPT Live Transcribe
OpenAI logoOpenAI
13.5%not measured4.9%620ms$0.0170
GPT Realtime Whisper
OpenAI logoOpenAI
16.8%not measured5.0%1.14s$0.0170
Qwen3-ASR Fast
Nari logoNari
12.0%3.3%3.8%33ms$0.0020
Qwen3-ASR
Nari logoNari
12.0%3.4%3.9%25ms$0.0010
GPT-4o-mini Transcribe
OpenAI logoOpenAI
20.6%2.7%5.7%822ms$0.0030
Grok Voice Transcribe 1.0
xAI logoxAI
13.7%2.8%37.2%1.13s$0.0033
Gradium ASR
Gradium logoGradium
24.2%8.4%11.6%1.13s$0.0104
Muse Voice Transcribe
meta logometa
14.2%not measured4.9%not measured$0.0030
Gemini 3.5 Transcribe Live
Google logoGoogle
13.1%not measurednot measured1.28s~$0.0090
MAI-Transcribe-2-Streaming
Azure logoAzure
not measurednot measurednot measurednot measurednot measured
ink-whisper
Cartesia logoCartesia
not measurednot measurednot measurednot measurednot measured
gemini-3.5-transcribeBatch only
Google logoGoogle
not measurednot measurednot measurednot measurednot measured
velma-2-stt-streaming
Modulate logoModulate
not measurednot measurednot measurednot measurednot measured
munsit
Munsit logoMunsit
not measurednot measurednot measurednot measurednot measured
munsit-en-ar
Munsit logoMunsit
not measurednot measurednot measurednot measurednot measured
default
Palabra logoPalabra
not measurednot measurednot measurednot measurednot measured

LLM21

ModelProviderId
DeepSeek-V4-Flash-0731
Baseten logoBaseten
93.16%558ms88%0%$0.26
gpt-4.1-mini
OpenAI logoOpenAI
92.63%480ms69%3%$1.60
gpt-5.4-mini
OpenAI logoOpenAI
91.75%390ms80%0%$4.50
gpt-4.1
OpenAI logoOpenAI
89.12%462ms78%10%$8.00
gemini-3.7-flash
Google logoGoogle
86.84%1.38s85%0%$3.75
Qwen3.8-27B
Cerebras logoCerebras
86.14%246ms86%0%$1.49
gemini-3.8-flash
Google logoGoogle
85.79%1.21s87%0%$3.75
DeepSeek-V4.1-Flash (Together)
Together logoTogether
84.21%157ms86%0%$1.20
gpt-5.6-terra
OpenAI logoOpenAI
83.51%580ms85%0%$12.00
Claude Sonnet 5
Anthropic logoAnthropic
83.33%1.47s89%0%$10.00
DeepSeek-V4.1-Flash (Baseten)
Baseten logoBaseten
82.19%458ms88%0%$1.20
gpt-oss-120b (Cerebras)
Cerebras logoCerebras
82.02%139ms53%0%$0.75
Claude Haiku 4.5
Anthropic logoAnthropic
81.67%391ms74%3%$5.00
gpt-oss-120b (Baseten)
Baseten logoBaseten
80.53%504ms73%0%$0.50
gpt-5.6-luna
OpenAI logoOpenAI
75.61%546ms78%0%$1.20
gpt-5.2
OpenAI logoOpenAI
75.61%664ms87%10%$14.00
Llama-3.3-70B
Together logoTogether
54.30%525ms57%0%$1.04
google/gemma-4-31B-it
DeepInfra logoDeepInfra
not measurednot measurednot measurednot measurednot measured
openai/gpt-oss-120b
Together logoTogether
not measurednot measurednot measurednot measurednot measured
Qwen/Qwen3.5-9B
Together logoTogether
not measurednot measurednot measurednot measurednot measured
zai-org/GLM-5.3-Flash
Together logoTogether
not measurednot measurednot measurednot measurednot measured

Text-to-speech35

ModelProviderId
eleven_v3_conversational
ElevenLabs logoElevenLabs
1590266ms$50.0
eleven_v4
ElevenLabs logoElevenLabs
~1708844ms$80.0
eleven_v4_turbo
ElevenLabs logoElevenLabs
~1670205ms$40.0
gemini-3.8-flash-tts
Google logoGoogle
~16421.03s~$15.0
gemini-3.8-flash-lite-tts
Google logoGoogle
~1639497ms~$10.0
flux
Deepgram logoDeepgram
~1550106ms$45.0
sonic-3.5
Cartesia logoCartesia
1574121ms$50.0
sonic-3.6
Cartesia logoCartesia
~1578120ms$50.0
tts-rt-v2
Soniox logoSoniox
~1606381ms~$13.0
simba-3.2
Speechify logoSpeechify
157385ms$10.0
inworld-tts-2
Inworld logoInworld
1561116ms$25.0
s2.1-pro
Fish Audio logoFish Audio
~1566185ms$15.0
lightning_v3.1
Smallest AI logoSmallest AI
1544173ms$17.5
grok-tts
xAI logoxAI
1495272ms$15.0
Gradium TTS
Gradium logoGradium
~1592105ms$57.8
gpt-4o-mini-tts
OpenAI logoOpenAI
1424691ms~$20.0
speech-2.8-hd
MiniMax logoMiniMax
1431294ms$100.0
palabra-tts-v1
Palabra logoPalabra
~152072ms$12.0
Qwen3-TTS Fast
Nari logoNari
~153664ms$10.0
Qwen3-TTS
Nari logoNari
~151075ms$5.0
paxa-tts-flash-v1
Paxa Labs logoPaxa Labs
~1325389ms$10.0
eleven_flash_v2_5
ElevenLabs logoElevenLabs
~1494not measured$50.0
eleven_multilingual_v2
ElevenLabs logoElevenLabs
~1472not measured$100.0
eleven_flash_v2
ElevenLabs logoElevenLabs
~1507not measured$50.0
sonic-3
Cartesia logoCartesia
~1551not measured$50.0
coda
Rime logoRime
~1535not measured$50.0
mistv3
Rime logoRime
~1428not measured$30.0
speech-2.6-turbo
MiniMax logoMiniMax
~1408not measured~$60.0
lightning_v3.1_pro
Smallest AI logoSmallest AI
~1447not measured$19.5
speech-2.8-turbo
MiniMax logoMiniMax
~1385not measured$60.0
MAI-Voice-2.1
Azure logoAzure
not measurednot measurednot measured
MAI-Voice-2.1-Flash
Azure logoAzure
not measurednot measurednot measured
eleven_v3
ElevenLabs logoElevenLabs
not measurednot measurednot measured
chirp-3-hd
Google logoGoogle
not measurednot measurednot measured
faseeh-v1-preview
Munsit logoMunsit
not measurednot measurednot measured

Speech-to-speech9

ModelProviderIdPriceas published
grok-voice-think-fast-2.0
xAI logoxAI
0.850.67820ms$0.08 · min
gpt-realtime-2.1-mini
OpenAI logoOpenAI
0.760.831.01s$10 / $20 · 1M tok
gemini-3.8-live
Google logoGoogle
0.870.831.21s$0.005 · min
gemini-3.8-live-extended-thinking
Google logoGoogle
0.830.83844ms$0.005 · min
gpt-live-1
OpenAI logoOpenAI
0.860.781.22s$0.05 · min
gpt-realtime-2
OpenAI logoOpenAI
0.670.871.10s$32 / $64 · 1M tok
gpt-realtime-2.1
OpenAI logoOpenAI
0.630.561.10s$32 / $64 · 1M tok
qwen3.8-livetranslate-flash-realtimeTranslation
Alibaba logoAlibaba
not measurednot measurednot measurednot measured
gpt-realtime-translateTranslation
OpenAI logoOpenAI
not measurednot measurednot measurednot measured

Published stacks

What benchmarks.speko.ai picks for each job, and the measurement that decided it.

Use case
Defaultdecided on 8.5% WER · 86% task success · Elo 1670
Speech-to-textMuse Voice Transcribemeta:muse-voice-transcribe-1.0
LLMQwen3.8-27Bcerebras:qwen-3.8-27b
Text-to-speecheleven_v4_turboelevenlabs:eleven_v4_turbo
Real-time phone agentdecided on 301ms · 246ms · 84ms
Speech-to-textMuse Voice Transcribemeta:muse-voice-transcribe-1.0
LLMQwen3.8-27Bcerebras:qwen-3.8-27b
Text-to-speechinworld-tts-2inworld:inworld-tts-2
Accuracy-criticaldecided on 7.2% WER · 93% task success · 0.90 robustness
Speech-to-textUniversal-3.5 Proassemblyai:universal-3-5-pro
LLMDeepSeek-V4-Flash-0731baseten:deepseek-ai/DeepSeek-V4-Flash-0731
Text-to-speecheleven_v4_turboelevenlabs:eleven_v4_turbo
Lowest costdecided on $0.0008/min · $0.0002/min · $0.0090/min
Speech-to-textVelma 2modulate:velma-2-stt-streaming-english-v2
LLMDeepSeek-V4-Flash-0731baseten:deepseek-ai/DeepSeek-V4-Flash-0731
Text-to-speechsimba-3.2speechify:simba-3.2