xAI Launches Grok Voice 3.0, Claims Top Speech-to-Speech Benchmark Scores
xAI launched Grok Voice 3.0, a new speech-to-speech voice model it calls “the world’s most intelligent speech-to-speech model,” now available in the Voice Agent API and powering voice mode in Grok Bot.
Details
- Benchmarks (Artificial Analysis, Oct 13, 2026): Grok Voice 3.0 scored 86.9 on the Speech-to-Speech Index (ranked #1), ahead of Grok Voice 2.0 (81.3), OpenAI’s GPT-Live-1/Astra Medium (83.2), and Google’s Gemini 3.8 Live Extended Thinking High (82.6)
- Other benchmark results: 98.5% on Big Bench Audio speech reasoning (vs. 97.2% / 90.1% / 97.7% for the same three comparison models); 1070 Elo on Speech Agent Arena human preference (vs. 1011 / 1048 / 990); 94.8% task success rate (vs. 94.6% / 87.4% / 89.1%); 97.0% on Full Duplex Bench for conversational turn-taking and interruption handling; 71.6% on τ-Voice support-call testing (noise, accents, interruptions), comparable to GPT-Live-1’s 74.5% and ahead of Gemini 3.8 Live’s 68.6%, while responding about 40% faster than GPT-Live-1
- Latency and cost: time to first audio under one second; average cost per Big Bench Audio task of $0.025, versus $0.030 for GPT-Live-1/Astra Medium and $0.055 for GPT-Realtime-2.1 High
- Enterprise benchmark: highest accuracy score (74%) and best combined accuracy/experience score on ServiceNow’s EVA-Bench
- Reported deployments: Tesla vehicles handle millions of voice conversations daily, letting drivers check calendars, read email via Connectors, or order coffee and book dinner hands-free through Grok Bot; Starlink call centers handle roughly 20,000 customer calls per day across multiple languages, resolving about 70% without a human and converting about one in four sales calls, with a single agent handling verification, troubleshooting, credits, and replacement shipping
- Pricing and rollout: available now in the Voice Agent API at $0.08/minute ($4.80/hour) — unchanged from Grok Voice 2.0; agents using the
grok-voice-latestalias were switched over automatically with no code changes required
What happened next
The release keeps xAI’s voice stack on a roughly quarterly upgrade cadence and leans heavily on third-party benchmark comparisons against OpenAI and Google’s latest voice models to stake a leadership claim, while the Tesla and Starlink deployment figures give the launch concrete, large-scale usage evidence beyond the benchmark charts.