Grok Voice 3.0 tops speech-to-speech benchmarks

Grok Voice 3.0’s meaningful advantage is operational: it combines benchmark-leading conversation quality with the latency and cost profile voice agents need in production.

• –Artificial Analysis testing cited by xAI gives it an 86.9 Speech-to-Speech Index score, ahead of Grok Voice 2.0, GPT-Live-1, and Gemini Live.
• –Full-duplex handling, interruption support, tool use, and multi-step reasoning make it better suited to support and sales agents than simple voice chat.
• –Starlink reportedly handles about 20,000 calls per weekday, with roughly 70% resolved without human intervention.
• –Developers get automatic migration through `grok-voice-latest`, reducing upgrade friction but limiting version control.
• –The claims are largely vendor-reported, so independent production testing across accents, failure recovery, and escalation quality still matters.