SpaceXSI, formerly xAI, released Grok Voice 3.0 on October 14, 2026, a speech-to-speech model that ranks first on Artificial Analysis's Speech-to-Speech Index with a score of 86.9, ahead of OpenAI's GPT-Live 1 at 83.2 and Google's Gemini 3.8 Live at 82.61,2. The model is available through SpaceXSI's Voice Agent API at $0.08 per minute, unchanged from Grok Voice 2.04.
Artificial Analysis, which published its rankings on October 13, 2026, bases the index on speech reasoning tests, task success rates, and listener votes in blind head-to-head calls. Grok Voice 3.0 scored 98.5% on Big Bench Audio speech reasoning, up from 97.2% for Grok Voice 2.0, and reached a Speech Agent Arena preference Elo of 1070, compared with 1011 for its predecessor. Full Duplex Bench, which grades how naturally a model handles pauses, turn-taking, and interruptions, gave the new model 97%, up from 95.1%3.
Where OpenAI still leads
OpenAI's GPT-Live 1 retained a narrow advantage on Tau-Voice, a benchmark built around simulated support calls with heavy accents and background noise, scoring 74.5% against Grok Voice 3.0's 71.6%. SpaceXSI countered that its model reached first audio roughly 40% faster on that test: 0.81 seconds versus 1.34 seconds for GPT-Live 1. Grok Voice 2.0 remains slightly quicker to start at 0.70 seconds.
On cost per task in Artificial Analysis's Big Bench Audio run, Grok Voice 3.0 came in at about 2.5 cents, compared with 3 cents for GPT-Live 1 and 5.5 cents for OpenAI's gpt-realtime-2.1.
Enterprise deployment and migration
SpaceXSI said Grok Voice powers millions of conversations a day across Tesla and SpaceX. The model also powers Grok Bot's voice mode and conversations in Tesla vehicles. Outside the Artificial Analysis suite, SpaceXSI reported a rounded 74% accuracy score on ServiceNow Inc's EVA-Bench, which evaluates enterprise voice agents.
Developers using the grok-voice-latest endpoint have been migrated automatically, and Grok Voice 2.0 has been deprecated, according to API release notes dated October 14. The release notes confirm real-time bidirectional audio and text streaming over WebSocket for assistants, phone agents, and interactive voice systems, with support for ephemeral tokens so client-side applications need not expose API keys.
ANALYSIS The per-task cost gap between Grok Voice 3.0 and gpt-realtime-2.1, roughly half the price, paired with a latency advantage over GPT-Live 1, gives SpaceXSI a concrete pitch to enterprise buyers running high-volume voice-agent workloads, even as OpenAI retains an edge on the accent-and-noise-heavy Tau-Voice test.