xAI launches Grok Voice 3.0 with under 1 second latency

xAI has released Grok Voice 3.0, a speech-to-speech model it calls the smartest in the world. The model set a new record in customer service call handling, with latency under 1 second and lower running costs than other leading models. It is now live via API at $0.08 per minute, the same price as the previous version.

In independent tests by Artificial Analysis, Grok Voice 3.0 topped the speech-to-speech rankings with a score of 86.9, and scored 97% on the full-duplex benchmark, which measures how naturally a model handles pauses, turn-taking and interruptions. Its tau-Voice score of 71.6% matches GPT-Live-1 (Astra), but with 40% lower latency.

The model powers millions of conversations daily inside Tesla vehicles, letting drivers check schedules, reply to emails or handle tasks like ordering food by voice alone. Starlink's customer service uses it to take 20,000 multilingual calls a day, with about 70% resolved without human help, and one in four sales calls closing before customers hang up.