xAI releases Grok Voice 3.0 speech model

xAI released the speech-to-speech model Grok Voice 3.0 on October 14, 2026, topping the Artificial Analysis Speech-to-Speech Index with a score of 86.9 and already processing approximately 20,000 calls per day in Starlink support.

Compiled by Sergey KostenchukPublished 2026-10-19Updated 2026-10-19

2026-10-19 Research OpenAI

🤖 xAI releases Grok Voice 3.0 speech model

Released on October 14, the speech-to-speech model listens and responds with voice, without intermediate text. According to Artificial Analysis, it topped the Speech-to-Speech Index with a score of 86.9, outperforming GPT-Live-1 and Gemini 3.8 Live; on the Full Duplex Bench it scored 97.0%, and on tau-Voice it scored 71.6% compared to 74.5% for GPT-Live-1.

🌍 Voice agents are moving from demos to mass deployment: the model is already working in Tesla cars and Starlink support—approximately 20,000 calls per day, with ~70% resolved without human intervention. The Voice Agent API price remains the same: 8 cents per minute, and agents on the grok-voice-latest alias have been updated automatically.

👤 For those building voice bots or considering AI operators instead of call centers, this is the strongest public result to date. The model does not support Russian, and the figures are currently from a single source.

Source 1: https://x.ai/news/grok-voice-3

Sources