Skip to content
AI.info

The Pulse

xAI Launches Grok Voice 3.0 at $0.08 a Minute

xAI launched Grok Voice 3.0 on October 14, 2026, as a speech-to-speech model for its Voice Agent API. The company says it costs $0.08 per minute and automatically replaces Grok Voice 2.0 for developers using the latest-model alias.

xAI Launches Grok Voice 3.0 at $0.08 a Minute

AI.info Team · Written with AI ·

“Grok Voice 3.0 sets a new Pareto frontier on EVA-Bench, ServiceNow’s open-source benchmark scoring voice agents on accuracy and experience across 200+ realistic enterprise scenarios. It pairs top accuracy with a natural, responsive conversation, setting a new standard for what enterprises can expect from AI on the front line of customer service.”

Nitin Aggarwal, Senior Director of Product Management, Agentic Evals and RAI, ServiceNow

xAI puts Grok Voice 3.0 in its API

xAI announced Grok Voice 3.0 on October 14, making the speech-to-speech model available through its Voice Agent API for $0.08 per minute, or $4.80 per hour. The company says the price matches Grok Voice 2.0 and that the API itself has not changed. Its release describes the model as delivering latency under one second, though that is xAI’s characterization rather than a guarantee for every application or network condition.

The launch targets developers building live voice agents for customer support and other tasks. xAI says the model can handle conversational interruptions and multi-step calls, and that it already powers millions of conversations a day across Tesla and SpaceX services. Those deployments give the launch a customer-service focus beyond a general-purpose voice assistant.

The benchmark lead comes with a specific scorecard

xAI’s announcement cites Artificial Analysis results dated October 13. Its comparison table gives Grok Voice 3.0 an overall Speech-to-Speech Index score of 86.9, against 81.3 for Grok Voice 2.0, 83.2 for GPT-Live-1 (Astra, Medium) and 82.6 for Gemini 3.8 Live (Extended Thinking, High). The index combines speech reasoning, agent performance, human preference and task success; it is a composite benchmark score, not a measure of every live customer interaction.

The same table reports a 97% score on Full Duplex Bench, which tests conversational timing such as pauses, turn-taking and interruptions. On τ-Voice, a customer-support task benchmark, Grok Voice 3.0 scores 71.6%; xAI says that is comparable with GPT-Live-1 (Astra) at 40% lower latency. Artificial Analysis describes τ-Voice as testing whether agents complete simulated customer-service tasks while using tools and policy documents. On ServiceNow’s EVA-Bench, xAI reports a 74% accuracy score and the top combined result for accuracy and customer experience.

Starlink supplies xAI’s clearest deployment figures

xAI says Starlink call centers use Grok Voice to answer 20,000 customer calls a day in multiple languages. The company reports that roughly 70% of those calls are resolved without a human agent, while one in four sales calls ends in a purchase before the caller hangs up. xAI does not provide a measurement period or a call-count denominator for the sales-conversion figure in its announcement.

The company says a single agent can verify customers, troubleshoot hardware, issue credits and ship replacement equipment. That sequence suggests the product is meant to do more than answer questions: it can carry out account and support workflows during a call. Tesla drivers, meanwhile, can use Grok Voice for hands-free requests, including calendar checks and email responses, or use Grok Bot for tasks such as ordering coffee or booking dinner.

Existing API users move over automatically

Developers using the grok-voice-latest alias have already been switched to version 3.0, according to xAI, with no code changes required. The company’s release notes also mark grok-voice-2.0 as deprecated and direct users to the new model slug or the alias. The same notes say the API remains unchanged and the price stays at $0.08 per minute.

The announcement pairs benchmark gains with claims about existing use at Tesla and Starlink, but the figures come from different kinds of evidence: controlled benchmark scores on one side and company-reported operational outcomes on the other. For developers, the immediate change is simpler to verify: the latest-model alias now points to Grok Voice 3.0, and the listed rate remains $0.08 per minute.

Sources

Explore

More articles