$ cat wiki/models/grok-voice-3.md
Grok Voice 3.0
Spec
| Attribute | Value |
|---|---|
| Developer | xAI; announcement branded SpaceXSI |
| Released | 2026-10-14 |
| Announced | 2026-10-14 |
| Context window | unknown |
| Pricing | $0.08 / min ($4.80 / hr) |
| License | unknown |
| Availability | Voice Agent API; grok-voice-latest automatically migrated |
| Specifications follow the release announcement; context capacity and license terms are not stated there. (source) |
Release Date
Announced and made available in the API on 2026-10-14. Existing users of the rolling alias were switched automatically; the vendor says pricing is unchanged from Grok Voice 2.0. (source)
Benchmarks
The vendor reproduces Artificial Analysis results dated 2026-10-13: AA Speech-to-Speech Index 86.9, Big Bench Audio 98.5%, Task Success Rate 94.8%, and Full Duplex Bench 97.0%. It separately reports τ-Voice 71.6% and EVA-Bench accuracy 74%. These are different metrics, not interchangeable success rates. The independent evaluator pages were not read in this capture. (source)
Use Cases
Speech-to-speech support agents and hands-free tool use. The vendor describes Starlink agents verifying customers, troubleshooting hardware, issuing credits and shipping replacement equipment; these are deployment claims rather than an independently audited reliability study. (source)
Compared To
The announcement quotes AA Speech-to-Speech Index 81.3 for Grok Voice 2.0, 83.2 for GPT-Live-1 (Astra, Medium), and 82.6 for Gemini 3.8 Live (Ext. Thinking, High). Preserve those effort settings and the evaluation date when comparing. Its stated 40% lower latency than GPT-Live-1 on τ-Voice is not a claim of 40% higher task success. (source)
Sources
- Release announcement, captured as the local snapshot.