Compare speech-to-speech models on real human calls
- ELO Rating11541069
- Reliability (Pass³)42.9%32.9%
- Hearing7571
- Turn Taking8172
- Speaking8783
- Task Handling8082
- Task Completion Rate34.8%36.2%
- Latency P501.60s2.27s
- Cost per Minute (USD)$0.080$0.055
- Measures won6 of 93 of 9
Any two of the 12 models, or any one against a cascade, on the same voice agent and the same calls.
Last updated Methodology by Dileep Chagam
By provider
The best model from each provider, metric by metric.
Every comparison
Grok Voice 3.0
GPT-Live 1 + GPT-6 Sol
Gemini 3.1 Flash Live Preview
GPT Realtime 2.1
Grok Voice Think Fast 2.0
Phonic-1
Azure Realtime
Gemini 3.8 Live Extended Thinking
Gemini 3.8 Live
GPT Realtime 2.1 Mini
Deepslate Opal
Amazon Nova 2 Sonic
CascadeNot ranked
Speech-to-speech model comparison FAQ
How do I compare speech-to-speech models?
Choose any two models to compare their ELO rating, reliability (pass³), the four skill families, calls fully resolved, latency and cost, all measured on the same recorded real human callers.
Which metrics matter most for a realtime voice model?
Start with ELO and reliability (pass³), then the skill family your calls depend on most: Hearing, Turn Taking, Speaking or Task Handling. Then latency and cost. A fast model is not useful if it cannot handle the same caller reliably.
Can I compare GPT Realtime with Gemini Live?
Yes. Select the tested GPT Realtime and Gemini Live versions to compare their reliability and latency on the same voice agent, prompt, tools and real human callers.
What is the difference between a native speech-to-speech model and a cascade?
A native model handles audio directly. A cascade connects speech-to-text, an LLM, and text-to-speech. The comparison pages test both as the same complete agent so their practical trade-offs are visible.
Test your voice agent the same way.
Cekura runs these simulated calls on your agent and scores every one.