x.ai Launches Grok Voice 3.0 Speech-to-Speech Model

x.ai has announced Grok Voice 3.0, describing it as an intelligent speech-to-speech model with sub-second latency and lower operational costs than competing frontier models. In independent testing by Artificial Analysis, Grok Voice 3.0 ranked first on the Speech-to-Speech Index with a score of 86.9 and scored 97.0% on Full Duplex Bench. In enterprise evaluations on ServiceNow's EVA-Bench, the model achieved a 74% accuracy score across more than 200 scenarios. The technology is deployed at scale across Tesla vehicles for in-car voice commands and handles 20,000 daily customer calls at Starlink support centers, resolving approximately 70% autonomously. Grok Voice 3.0 is offered via the Speech to Speech API at $0.08 per minute ($4.80 per hour).

Key takeaways

• In Artificial Analysis benchmarks, Grok Voice 3.0 scored 86.9 on the Speech-to-Speech Index, 97.0% on Full Duplex Bench, and 71.6% on τ-Voice at 40% lower latency than GPT-Live-1.
• On ServiceNow's EVA-Bench evaluating 200+ realistic enterprise scenarios, Grok Voice 3.0 achieved a top accuracy score of 74%.
• The model handles millions of daily Tesla vehicle interactions and 20,000 daily Starlink support calls, autonomously resolving about 70% of support issues.
• Grok Voice 3.0 is available in the Speech to Speech API at $0.08 per minute ($4.80 per hour), matching the price of Grok Voice 2.0.