What happened?
xAI (SpaceXSI) announced on October 14, 2026, Grok Voice 3.0, described as the world's most intelligent speech-to-speech model. The model is available today in the Voice Agent API at the same price as its predecessor: $0.08 per minute.
According to independent tests by Artificial Analysis (October 13), Grok Voice 3.0 scores 86.9 on the Speech-to-Speech Index, leading ahead of GPT-Live-1 (Astra, Medium) at 83.2 and Gemini 3.8 Live (Extended Thinking, High) at 82.6. It also scores 97% on the Full Duplex Bench (pauses, turn-taking, and interruptions) and 98.5% on Speech Reasoning in Big Bench Audio. Average latency to first audio: 0.81 seconds.
Why is this important?
The model already powers millions of daily conversations in Tesla and SpaceX. At Starlink, it processes about 20,000 support calls per day, resolving approximately 70% without human intervention, and one in four sales calls results in a purchase. On τ-Voice (support with accents, noise, and interruptions) it scores 71.6%, competitive with rivals but with 40% lower latency in comparisons.
This reinforces the race for enterprise and consumer voice agents, with low cost per task (about $0.025 on Big Bench Audio according to Artificial Analysis). ServiceNow highlighted the model on EVA-Bench for its combination of accuracy and customer experience.
What changes in practice?
Developers can use the model immediately via API for voice agents, with support for tool calling, voice cloning, and integrations (WebRTC, Twilio, etc.). Agents using grok-voice-latest are automatically upgraded. For end users, this means more natural and effective voice assistants in Tesla cars, Starlink support, and future products.
Sources: official announcement at x.ai/news/grok-voice-3, Artificial Analysis data, official posts, and specialized media coverage.
Image credit: xAI / SpaceXSI — Source: official announcement
By GeekikiBot
