xAI has released Grok Voice 3.0, which achieved the top performance in the Speech-to-Speech benchmark.

Beating AI reports that xAI has launched the new voice-to-voice model, Grok Voice 3.0. According to official specifications, this model achieved an outstanding score of 86.9 in the Independent Testing conducted by Artificial Analysis for the Speech-to-Speech Index. In comparison, Grok Voice 2.0 scored 81.3, while GPT-Live-1 scored 83.2, and Gemini 3.8 Live scored 82.6. The model also features an end-to-end latency of less than 1 second.<br /> <br /> Currently, this model is available through the Voice Agent API. The pricing is $0.08 per minute, the same as previous versions. Using the grok-voice-latest agent will automatically switch between this model and others.<br /> <br /> Officials stated that Grok Voice handles millions of conversations daily for companies like Tesla and SpaceX. The Starlink customer service center receives approximately 20,000 calls per day, of which about 70% can be resolved without human intervention. Approximately one-fourth of sales calls are completed before being hung up. On the ServiceNow EVA-Bench test platform, the model achieves an accuracy rate of 74%, compared to 71.6% for τ-Voice. Officials claim that Grok Voice’s performance is comparable to GPT-Live-1, with a latency that is about 40% lower.