SpaceXAI released Grok Voice 3.0 on 14 October. In the Artificial Analysis index the model scores 86.9 against 83.2 for GPT-Live-1, in a table published by SpaceXAI itself. The price is the same as Grok Voice 2.0.
- 14 October 2026: SpaceXAI announced Grok Voice 3.0, a speech-to-speech model it calls the most intelligent in the world. It is available in the Voice Agent API.
- Per Artificial Analysis, as cited by SpaceXAI: 86.9 in the Speech-to-Speech Index against 83.2 for GPT-Live-1 and 82.6 for Gemini 3.8 Live. On τ-Voice GPT-Live-1 scores higher: 74.5% against 71.6%.
- Price: $0.08 a minute ($4.80 an hour), the same as Grok Voice 2.0. Agents on grok-voice-latest were switched over automatically.
$0.08 a minute. That is what Grok Voice 2.0 cost and what 3.0 costs. The price is the line nobody will quote. They'll quote the other one: SpaceXAI calls Grok Voice 3.0 the most intelligent speech-to-speech model in the world.
The index belongs to an outside firm, and that weighs more than the usual in-house number. But the table is published on SpaceXAI's page, and it holds only Grok Voice 3.0, its predecessor, GPT-Live-1 and Gemini 3.8 Live. Which other models are in the index and where they stand, the page doesn't say. In the table, GPT-Live-1 is second with 83.2.
On τ-Voice the model doesn't lead. The test measures how agents handle support calls: Grok Voice 3.0 scores 71.6%, GPT-Live-1 74.5%. SpaceXAI presents it as comparable and adds the lower latency. “Comparable” is its word, and the 20,000 calls a day at Starlink are its number.
If you use grok-voice-latest, you're already on 3.0: SpaceXAI switched those agents over automatically. The price is the same. The behavior may not be. Run your test calls again.