xAI released Grok Voice 3.0 on October 14, 2026, calling it “the world’s most intelligent speech-to-speech model.” It is available in xAI’s Voice Agent API, where agents using grok-voice-latest switched over automatically, at $0.08 per minute ($4.80 an hour), the same price as Grok Voice 2.0.
On figures xAI attributes mostly to Artificial Analysis, Grok Voice 3.0 scores 86.9 on the Speech-to-Speech Index (Grok Voice 2.0: 81.3; GPT-Live-1 Astra: 83.2; Gemini 3.8 Live: 82.6), 98.5 percent on Big Bench Audio and 97.0 percent on Full Duplex Bench. On the tau-Voice agent benchmark it scores 71.6 percent, slightly behind GPT-Live-1 at 74.5 percent, though xAI says it does so with 40 percent lower latency. xAI reports sub-second latency and 74 percent accuracy on ServiceNow’s EVA-Bench.
xAI says the model handles millions of conversations a day across Tesla vehicles and SpaceX, and that Starlink call centers, at about 20,000 calls a day, resolve roughly 70 percent without a human.
Why it matters: voice agents for customer support became a direct competition among frontier labs in 2026, and the comparison is now run on shared third-party indexes and cost per task rather than demos.
What it does not show: the benchmark table is selected and published by xAI, it does not lead on every row, and the Starlink resolution and sales figures are xAI’s own operational claims about a sister company.