Grok Voice 3.0 is SpaceXAI’s speech-to-speech successor to Grok Voice 2.0, offered through its Realtime API for $0.08 per minute, according to Artificial Analysis. The evaluation used the Rex voice.

Grok Voice 3.0 is SpaceXAI’s speech-to-speech successor to Grok Voice 2.0, offered through its Realtime API for $0.08 per minute, according to Artificial Analysis. The evaluation used the Rex voice.

Artificial Analysis reports a leading Speech-to-Speech Index score of 86.9, ahead of GPT-Live-1 (Astra, medium) at 83.2 and Gemini 3.8 Live Extended Thinking (High) at 82.6. Its index combines audio reasoning, agentic performance, arena preference, and task success.

The component results are more mixed: Grok Voice 3.0 scored #1 on task success at 94.8%, #2 on Tau Voice at 71.6%, and 98.5% on Big Bench Audio reasoning. It placed #3 in arena preference at Elo 1070. Its Big Bench Audio time to first audio was 0.81 seconds, faster than the listed GPT and Gemini comparisons, but slower than Grok Voice 2.0 at 0.70 seconds.

The post lists input audio cost at $4.80 per hour, equal to Grok Voice 2.0. It is cheaper than GPT-Live-1 (Astra, medium) at $5.83 and GPT-Realtime-2.1 High at $10.75, but costs more than GPT-Live-1 (Sol, low) at $4.47 and Gemini 3.8 Live Extended Thinking (High) at $3.50.

Claims that it handles millions of daily Tesla conversations and supports hands-free tasks such as ordering coffee come from XFreeze’s post; the benchmark post does not establish those deployment claims.