SpaceXSI debuts Grok Voice 3.0, saying it tops Artificial Analysis' speech-to-speech index, responds in under a second, and costs less than rivals, at $0.08/min (SpaceXSI)

SpaceX has unveiled Grok Voice 3.0, an advanced speech-to-speech model that outperforms other AI in customer support calls, offers lightning-fast response times, and is priced lower than its rivals. The model, which powers millions of daily conversations across Tesla and SpaceX, is now accessible via the API at the same cost as its predecessor.

Artificial Analysis' independent testing has crowned Grok Voice 3.0 as the top-ranked model on the Speech-to-Speech Index, with a score of 86.9. The index evaluates speech reasoning, τ-Voice, and preference and task success from blind head-to-head calls. Grok Voice 3.0 also excels in Full Duplex Bench, a metric for handling pauses, turn-taking, and interruptions, scoring 97% and outpacing GPT-Live-1 (Astra) by a significant margin while delivering 40% lower latency.

Additionally, Grok Voice 3.0 boasts a commendable 71.6% accuracy in managing diverse accents, background noise, and interruptions, comparable to GPT-Live-1 (Astra) but with 40% reduced latency. The model also excels in enterprise voice agent benchmarks, achieving the highest accuracy score of 74% and outperforming competitors in terms of both accuracy and customer experience.

Grok Voice 3.0 is already being used in Tesla vehicles, where drivers can utilize Connectors for simple requests and Grok Bot for more complex tasks such as ordering coffee or booking dinner—all hands-free. The model also powers Grok Bot's voice mode and has been successfully deployed in Starlink call centers, handling over 20,000 customer calls daily in multiple languages, with 70% resolution without human intervention.