What it does
Grok Voice 3.0 is SpaceXAI's speech-to-speech model, announced on 14 October 2026 for voice agents such as customer support callers. It is available through the Voice Agent API, and the alias grok-voice-latest points to the current version, so agents using it were switched over automatically. The announcement reports Artificial Analysis testing in which Grok Voice 3.0 scores 86.9 on its Speech-to-Speech Index and 97% on Full Duplex Bench, which measures handling of pauses, turn-taking and interruptions. It also cites a 74% accuracy score on ServiceNow's EVA-Bench for enterprise voice agents. These are figures published by the vendor and the benchmarks it cites, not checked here. The post says the model handles millions of conversations a day in Tesla vehicles and powers Grok Bot's voice mode, and that Starlink call centres use it for 20,000 calls a day across several languages, with roughly 70% resolved without a human. Billing is $0.08 per minute, the same as Grok Voice 2.0.
Capabilities
- Speech-to-speech model for conversational voice agents
- Time to first audio under one second, according to the company
- Handles pauses, turn-taking and interruptions
- Used across multiple languages in Starlink call centres
- A single agent can verify customers, troubleshoot hardware, issue credits and ship replacements
- The grok-voice-latest alias tracks the current version
When to use it
- Building phone or in-vehicle voice agents that need quick, interruptible conversation
- Automating customer-support calls that involve actions such as credits or replacement shipments
- Comparing voice-agent models on cost per minute and response time
