SpaceXSI has released Grok Voice 3.0, its latest speech-to-speech model. In a post dated 14 October 2026, the company says the model is available now in its Voice Agent API at the same price as Grok Voice 2.0, and presents benchmark scores from Artificial Analysis in which Grok Voice 3.0 ranks first on the overall Speech-to-Speech Index. OpenAI still leads on τ-Voice.
Benchmarks, as SpaceXSI presents them
SpaceXSI cites Artificial Analysis testing dated 13 October 2026. On the overall Artificial Analysis Speech-to-Speech Index, which the company says combines speech reasoning, τ-Voice, and preference and task success from blind head-to-head calls, the scores it lists are:
| Model | AA Speech-to-Speech Index |
|---|---|
| Grok Voice 3.0 | 86.9 |
| GPT-Live-1 (Astra, Medium) | 83.2 |
| Gemini 3.8 Live (Ext. Thinking, High) | 82.6 |
| Grok Voice 2.0 | 81.3 |
Other figures SpaceXSI attributes to Artificial Analysis for Grok Voice 3.0 include 98.5% on Speech Reasoning (Big Bench Audio), an Elo of 1070 on Human Preference (Speech Agent Arena), 94.8% task completion, and 97.0% on Full Duplex Bench, which measures pauses, turn-taking and interruptions.
On τ-Voice, which Artificial Analysis uses for support-call handling with accents, background noise and interruptions, SpaceXSI says Grok Voice 3.0 scores 71.6%. Its own chart on the same page labels GPT-Live-1 (Astra, Medium) at 74.5%, so OpenAI still leads on that measure. SpaceXSI says Grok Voice 3.0 has about 40% lower latency and shows “71.6% at 0.81s”.
On cost per task in Artificial Analysis’s Big Bench Audio run, SpaceXSI’s chart shows Grok Voice 3.0 at $0.025, GPT-Live-1 at $0.030 and GPT-Realtime-2.1 at $0.055.
Separately, SpaceXSI says Grok Voice 3.0 scores 74% accuracy on EVA-Bench, ServiceNow’s enterprise voice-agent benchmark, and quotes ServiceNow’s Nitin Aggarwal calling it a new “Pareto frontier” on that test. That accuracy claim is presented as SpaceXSI’s reading of ServiceNow’s benchmark.
Price and availability
Grok Voice 3.0 is available in the Voice Agent API at $0.08 per minute ($4.80 per hour), the same price as Grok Voice 2.0, according to SpaceXSI. Agents using grok-voice-latest have already been switched over automatically, it says.
How SpaceXSI says it is used
SpaceXSI says Grok Voice powers millions of conversations a day across Tesla and SpaceX. It claims Starlink call centres use Grok Voice for 20,000 customer calls a day, with roughly 70% resolved with no human in the loop, and that one in four sales calls ends in a purchase. Those Tesla and Starlink figures are SpaceXSI’s own claims.
What this means
- The launch and API price are confirmed by SpaceXSI. Same price as the previous version, live today.
- The headline ranking is Artificial Analysis as cited by SpaceXSI. TSN has not re-run the tests.
- OpenAI still leads on τ-Voice. Grok Voice 3.0 is ahead on the overall Speech-to-Speech Index in the figures SpaceXSI published, but not on that support-call score.
Sources
- SpaceXSI / x.ai, “Introducing Grok Voice 3.0”, 14 October 2026. https://x.ai/news/grok-voice-3
