News

Releases & Models

CONFIRMED: xAI's Grok Voice 3.0 takes the lead in independent voice benchmark without a price hike

xAI officially launched Grok Voice 3.0 with a rare piece of independent validation: an Artificial Analysis benchmark puts the model at the top of conversational voice assistants, with sub-second response times and an unchanged price.

Ilustracao editorial representando onda de voz da IA Grok Voice 3.0 da xAI

xAI launches Grok Voice 3.0 with numbers to back the promise

xAI officially announced Grok Voice 3.0, the new version of its voice model, already available via API for developers under the grok-voice-latest identifier. The launch comes with something rare in the AI market: independent validation. According to the Speech-to-Speech Index, maintained by Artificial Analysis, an organization that tests models in isolation with no commercial ties to vendors, Grok Voice 3.0 scored 86.9, beating the previous version's 81.3 and taking the lead among evaluated conversational speech models.

Speed you can feel in the conversation

The new model's response time dropped below one second, at 0.81 second according to Artificial Analysis. In practice, that means less of the awkward pause between a user's speech and the assistant's reply, one of the biggest experience gaps in voice assistants to date. xAI claims customer service performance also improved over the previous generation, though that specific point relies on the company's internal metrics, not the third party benchmark.

Price stayed flat, upgrade is automatic

Despite the performance jump, xAI kept the price at $0.08 per minute, the equivalent of $4.80 per hour, the same rate charged for Grok Voice 2.0. Anyone already running agents configured with the grok-voice-latest identifier gets the update automatically, with no code or configuration changes needed. That lowers adoption friction for teams already building voice products on xAI's infrastructure.

Already in heavy production before the public announcement

According to xAI, Grok Voice already handles millions of conversations a day inside Tesla vehicles, responding to requests like checking a calendar or replying to an email completely hands free, and also powers Grok Bot for longer tasks such as ordering coffee or booking dinner. At Starlink's call centers, the model answers around 20,000 calls a day across multiple languages, with 70% resolved with no human in the loop and one in four sales calls ending in a purchase before the caller hangs up. Those numbers come from xAI itself and have not been independently audited, but they point to the real usage scale that preceded the announcement.

What still needs confirming

The Artificial Analysis benchmark is independent, but it measures latency and speech quality in controlled scenarios that do not necessarily reproduce the real support complexity xAI cites. It remains to be seen how competitors like OpenAI and Google respond in the next evaluation cycle, and whether the performance gain holds up under massive use outside the test environment. For now, what is confirmed is the launch, the benchmark result and the unchanged price; the impact on third party production use is still an open question.

Sources

xAI: Introducing Grok Voice 3.0; Technobezz: xAI Launches Grok Voice 3.0 at $0.08 Per Minute; Hermes AI News: Grok Voice 3.0 Tops Speech Benchmark.