Grok Voice 3.0 has taken the top spot on the Artificial Analysis Speech-to-Speech Index, giving SpaceXAI’s voice model the highest score on the ranking. It posted an Index score of 86.9, ahead of every model Artificial Analysis currently ranks.
The model is not only a benchmark result. Starlink uses Grok Voice to handle customer service calls, where it answers 20,000 calls every day across multiple languages. Roughly 70% of those calls end without a human joining the conversation.
A top-ranked model built for live conversations
Grok Voice 3.0 is SpaceXAI’s flagship realtime voice model. Developers can access it as grok-voice-3.0 through the /v1/realtime WebSocket endpoint, and the model is compatible with the OpenAI Realtime API.
That combination gives the model a clear position: it is designed to listen and respond in live voice exchanges rather than handle text alone. Its time to first audio is 0.81 seconds, a figure that points to the short gap between a request and the model’s first spoken response.
Grok Voice 3.0 runs on SpaceXAI’s hosted Voice Agent API only. Its listed cost is $0.025 per task, which places a clear price on each completed interaction for teams using the service.
The dates attached to the update are Oct. 11, 2026, and October 14, 2026. By October 14, 2026, Grok Voice 3.0 ranked first on the Artificial Analysis Speech-to-Speech Index with its 86.9 score.
Starlink puts the voice agent to work
Starlink’s use of the model shows what those benchmark results look like in a working customer service operation. Its call centers use Grok Voice to answer 20,000 customer calls every day across multiple languages, with roughly 70% of calls resolved without a human in the loop.
That resolution rate covers more than simple questions. A single agent built on Grok Voice 3.0 verifies customers, troubleshoots hardware, issues credits, and ships replacement equipment. Those tasks give the voice agent responsibility across several parts of a customer service call.
The system also handles sales conversations. One in four sales calls at Starlink ends in a purchase before the caller hangs up, which equals a 25% purchase rate for those calls.
That figure sits beside the 70% resolution rate, but the two numbers describe different outcomes. The first measures calls completed without human help, while the second measures sales calls that end in a purchase. Together, they show the model handling both support and sales work.
Why the ranking matters
An 86.9 score gives Grok Voice 3.0 the highest Index score among the models Artificial Analysis ranks. The 0.81-second time to first audio adds a response-speed measure, while the $0.025 per-task price gives users a cost measure to consider.
Those numbers do not stand alone. The model is already connected to a large call workload through Starlink, where it answers 20,000 calls each day and works across multiple languages. That makes its role easier to understand than a leaderboard position by itself.
The hosted-only setup also defines how developers use Grok Voice 3.0. Access runs through SpaceXAI’s Voice Agent API, with the model exposed as grok-voice-3.0 on the /v1/realtime WebSocket endpoint. Compatibility with the OpenAI Realtime API gives the model another point of connection for realtime voice applications.
For customers, the practical test is simple: can the agent complete the call? Starlink’s figures provide a direct answer for its operation. Roughly 70% of calls need no human in the loop, while a single agent can verify a customer, troubleshoot hardware, issue credits, and ship replacement equipment.
Grok Voice 3.0 now combines the highest Artificial Analysis Speech-to-Speech Index score with a live customer service role. Its 86.9 score marks the benchmark result, and Starlink’s 20,000 daily calls show where SpaceXAI is putting the model to work.
Based on
