SpaceXSI announced Grok Voice 3.0 on October 14, 2026, calling it "the world's most intelligent speech-to-speech model." The company says the new model sets a record for resolving customer support calls and keeps latency under one second, while costing less than other frontier voice models. It is available today through the Voice Agent API, reports Techmeme.
The claim rests on independent testing by Artificial Analysis, where Grok Voice 3.0 ranks number one on the Speech-to-Speech Index with a score of 86.9. That index combines speech reasoning, τ-Voice and blind head-to-head call preferences. It is a strong marketing position for SpaceXSI, but the numbers come from third-party evaluation rather than internal testing alone.
Grok Voice 3.0 benchmarks and pricing
The published results are mixed but mostly favorable. On τ-Voice, which tests support calls with varied accents, background noise and interruptions, Grok Voice 3.0 scores 71.6% — below GPT-Live-1 (Astra, Medium) at 74.5% but above Gemini 3.8 Live (Ext. Thinking, High) at 68.6% and far above Grok Voice 2.0 at 56.5%. SpaceXSI argues the comparison matters most at 40% lower latency than GPT-Live-1.
Cost is the other headline. Artificial Analysis measures the average cost to complete a task on its Big Bench Audio run: Grok Voice 3.0 comes in at $0.025, against $0.030 for GPT-Live-1 (Astra, Medium) and $0.055 for GPT-Realtime-2.1 (High). On EVA-Bench, ServiceNow's benchmark for enterprise voice agents, the model reaches 74% accuracy, the highest score in that test according to SpaceXSI.
- Speech-to-Speech Index score of 86.9, ranked #1 by Artificial Analysis
- 97% on Full Duplex Bench for pauses, turn-taking and interruptions
