Grok Voice 3.0 launches with a leading speech-to-speech index score
The new voice model leads Artificial Analysis’s composite index at the time checked; API pricing starts at $0.08 per minute.
- Event or announcement
- Sources checked
What happened
SpaceXSI announced Grok Voice 3.0 on October 14, 2026. The Voice Agent API’s grok-voice-latest alias now selects version 3.0; the documentation also allows the explicit grok-voice-3.0 model name.
Read the benchmark narrowly
Artificial Analysis’s table, checked October 15, gives Grok Voice 3.0 the highest listed composite Speech-to-Speech Index score, 86.9. Comparators include GPT-Live-1 (Astra, medium) at 83.2 and Gemini 3.8 Live Extended Thinking (High) at 82.6.
Its 97.0% conversational-dynamics result and 71.6% agentic-performance result are benchmark measurements, not guaranteed customer-call outcomes. The table reports 0.81 seconds to first audio on its test; that is not a promise for every deployment. Composite leadership does not mean leading every individual metric.
Costs and a safe trial
Official pricing lists USD $0.08 per minute ($4.80 per hour) and a $0.004-per-text-input rate. Budget for applicable tool and deployment costs rather than treating the hourly figure as an all-in phone service.
Editorial suggestion: Pin the version during evaluation. Test interruptions, noisy audio, accents, tool failures and human handoffs using consented, non-sensitive calls. Keep permanent API credentials on the server; the documentation recommends short-lived tokens for client apps.
Read the sources
- SpaceXSI launch announcement (opens in a new tab)
- Artificial Analysis speech-to-speech benchmarks (opens in a new tab)
- SpaceXSI API pricing (opens in a new tab)
- Speech-to-speech API documentation (opens in a new tab)
This is a source-based summary. Product performance and account access have not been independently tested here. Practical suggestions are editorial.