Grok Voice 3.0 tops the voice AI rankings, and your phone is next

On Wednesday, October 14, SpaceXAI released Grok Voice 3.0, which it calls the world's most intelligent speech-to-speech model. Speech-to-speech means the model listens and answers in audio, instead of chaining a transcriber, a chatbot and a synthetic voice together. Fewer handoffs, faster replies: the company says Grok Voice 3.0 responds in under a second, per its announcement.

It costs $0.08 a minute in the Voice Agent API, or $4.80 an hour, the same price as version 2.0. Apps that use the grok-voice-latest alias switched over on their own, and Grok Voice 2.0 is now deprecated, per SpaceXAI's release notes.

That is roughly the price of one fancy coffee per hour, for an agent that never takes a coffee break.

Artificial Analysis, an independent firm that benchmarks AI models, puts Grok Voice 3.0 first on its Speech to Speech Index with a score of 86.9. OpenAI's GPT-Live-1 (Astra, medium) scores 83.2 and Google's Gemini 3.8 Live with extended thinking scores 82.6, per the Artificial Analysis leaderboard. The index blends four measures: reasoning out loud, finishing agent tasks, overall task success, and which voice people prefer in blind head-to-head calls.

Read the fine print, though. On Tau-Voice, a test of support calls with accents, background noise and interruptions, Grok Voice 3.0 scored 71.6%, a little behind GPT-Live-1 at 74.5%, though SpaceXAI says it answers with about 40% lower latency, per the announcement. The company also chose which charts to show, so the independent leaderboard is the number to lean on.

So it won the decathlon, not every event. That is still a medal worth noticing.

SpaceXAI says Grok Voice runs Starlink's call centers, which handle about 20,000 customer calls a day in several languages. About 70% are resolved without a human, and roughly one in four sales calls ends in a purchase before the caller hangs up, the company said. It adds that the model handles millions of conversations a day in Tesla cars.

Those are the company's own numbers from businesses tied to Elon Musk, with no outside audit. Still, they describe the job most businesses actually need done on the phone: confirm who is calling, fix the simple problem, and make the sale when the caller is ready.

It is tempting to assume everyone books online now. The data disagrees. In a national survey of 3,661 US adults run in late 2024, 56.4% said phone calls were their main way to schedule medical appointments, far ahead of patient portals at 19.7%, per a study in Health Affairs Scholar. Nearly three in four (72.1%) had booked by phone at least once that year.

And the call usually starts on a phone screen. In BrightLocal's July 2026 survey of 1,227 US consumers, 73% began their latest local business search on a mobile phone and 75% used more than one channel before deciding, per BrightLocal. Only 8% started with an AI tool, and 43% of those went back to Google to double-check.

So the path looks like this: a search on a phone, a quick look at reviews and listings, then a tap on your number. Voice AI changes that last step, the one where money changes hands.

The pitch is easy to love: no missed calls, no hold music, answers at 2 a.m. The risk is just as clear. A caller who wanted a person and got a loop of polite confusion will not call twice, and might write a review about it.

Start narrow. Let the agent handle the questions your team answers twenty times a day, like hours, parking, services and booking. Give callers a fast, obvious way to reach a person, and have the agent say up front that it is an AI assistant. Callers forgive a robot that is honest and quick. They do not forgive one pretending to be Brenda from the front desk.

Then measure it like a new hire, not a gadget. Track answered calls, bookings, transfers to a person and complaints for the first 30 days, and compare them with the month before.

People will keep calling. What changes is who picks up. The businesses that come out ahead will be the ones with consistent facts everywhere and a fast handoff to a real person, and you can fix both in a week, no API key required.

If you want help getting your listings, call flow and follow-up ready for AI, our founders will walk you through it. Talk to them below. See you in the next edition.