Automation · Oct 15, 2026
AI customer support calls: what Grok Voice 3.0 shows about voice agents in 2026
On October 14, xAI announced Grok Voice 3.0, a speech-to-speech model it calls the most intelligent voice model available. It ranks first on Artificial Analysis's independent Speech-to-Speech Index with a score of 86.9, ahead of OpenAI's GPT-Live-1, which launched in September.
The benchmark is interesting. The real-world example xAI shared is more useful.
The Starlink example
According to xAI, Starlink call centers use Grok Voice to answer 20,000 customer calls a day across multiple languages. xAI says roughly 70% of those calls are resolved with no human involved, and that a single agent verifies customers, troubleshoots hardware, issues credits, and ships replacement equipment.
That is a company-owned example from a related company, so treat it as a best case. But look at what the agent does: verify, troubleshoot, credit, and ship. Every one of those is a defined action connected to a real system. That is the part that makes voice agents work.
The metrics that matter for a business
When owners ask me whether a voice agent is "good," I steer them away from benchmark scores and toward a handful of practical measures. xAI's announcement happens to touch on most of them.
Resolution rate. What share of calls end with the caller's problem solved and no human needed? This is the number that determines whether the agent saves time.
Latency. How long does the caller wait after they stop talking? xAI says Grok Voice 3.0 responds in under one second. Long pauses make callers talk over the agent and hang up.
Turn-taking. Can the agent handle interruptions and pauses naturally? xAI reports 97% on Full Duplex Bench, a test of exactly that.
Handoff quality. When the agent cannot help, does it pass the caller to a person with the context, or make them start over?
Cost per resolved call. Per-minute pricing is easy to compare, but cost per resolved call is what matters. xAI lists its price at $0.08 per minute, the same as the previous version.
How the market has changed since September
In September, OpenAI brought GPT-Live-1 to its API with phone support. In October, xAI released Grok Voice 3.0. Both now support real phone lines, handle interruptions, and connect to tools. Voice agents have moved from novelty to a competitive market in a matter of weeks.
For a business, competition means better quality and stable or falling prices. It also means you should avoid building your phone system so tightly around one provider that you cannot switch.
Where to start
The businesses that do well with voice agents start narrow. Pick one call type that is common, repetitive, and connected to a system you already have. Appointment changes, order status, service area questions, and after-hours intake are good candidates.
Then measure for a month. Read transcripts. Track resolution rate and handoffs. Expand only when the numbers hold up.
What this means for your business
- Voice agents are now good enough that large companies run high-volume support on them.
- Judge them on resolution rate, latency, handoffs, and cost per resolved call, not leaderboard scores.
- The value comes from connecting the agent to your scheduling, order, or job systems.
- Start with one call type, measure, then expand.
If you are curious whether a voice agent could take some pressure off your phones, we can help you look at one connected to the systems you already use.
