What happened
SpaceXAI announced Grok Voice 3.0, its flagship real-time voice model. It is described as a speech-to-speech system that listens, reasons, calls tools and speaks. It took the top spot on the Artificial Analysis Speech-to-Speech Index with a score of 86.9.
How it works
The model is available only through SpaceXAI's hosted Voice Agent API; there are no open weights, local or quantized variants. It is exposed as grok-voice-3.0 on the /v1/realtime WebSocket endpoint, and the grok-voice-latest alias already points to it. Reasoning is on by default and can be turned off for faster turns. Server-side voice activity detection handles barge-in.
Benchmarks
The AA Speech-to-Speech Index is an equal-weighted average of four components. Grok Voice 3.0 scores 86.9, ahead of GPT-Live-1 (Astra, medium) at 83.2 and Gemini 3.8 Live Extended Thinking at 82.6. Time to first audio is 0.81 seconds, well ahead of GPT-Live-1's 1.34 seconds. On τ-Voice agentic tasks, however, it trails GPT-Live-1 at 71.6 versus 74.5.
Price and access
Pricing stays at $0.08 per minute and $4.80 per hour, the same as Grok Voice 2.0. The model supports 28 built-in voices, voice cloning, 20 locales and five tool types: files, web, X, MCP and custom functions.
