Image to help understand the article
New voice model reaches developers immediately
xAI unveiled Grok Voice 3.0 on Oct. 14, 2026, offering its latest artificial intelligence voice model through the company’s Voice Agent API. The company says the system can begin producing a spoken response in less than one second.
Access costs 8 cents per minute, or $4.80 per hour, unchanged from Grok Voice 2.0. Developers already using the model identifier grok-voice-latest were automatically moved to the new version without needing to change their code.
xAI reports gains across voice benchmarks
According to figures released by xAI, Grok Voice 3.0 scored 86.9 on the AA Speech-to-Speech Index, compared with 81.3 for Grok Voice 2.0, 83.2 for OpenAI’s GPT-Live-1 and 82.6 for Gemini 3.8 Live.
Artificial Analysis, the independent organization that operates the index, says it gives equal weight to tests of spoken reasoning, conversational dynamics and an AI agent’s ability to complete tasks. However, an index description published by the organization on June 23 did not include results for Grok Voice 3.0.
In xAI’s reported results, the model scored 98.5% on Big Bench Audio’s speech-reasoning test and 97% on Full Duplex Bench, which measures simultaneous conversational performance. It also posted a Speech Agent Arena Elo rating of 1070 and a task success rate of 94.8%.
xAI estimated that each Big Bench Audio task cost about 2.5 cents with Grok Voice 3.0, compared with 3 cents for GPT-Live-1 and 5.5 cents for GPT-Realtime-2.1.
OpenAI model leads in one task-focused evaluation
Grok Voice 3.0 did not lead every test. GPT-Live-1 scored 74.5% on the τ-Voice benchmark, ahead of Grok Voice 3.0 at 71.6%. Gemini 3.8 Live scored 68.6%, while Grok Voice 2.0 reached 56.5%.
xAI said its new model has roughly 40% lower latency than GPT-Live-1, describing the speed advantage as a trade-off against the OpenAI model’s higher τ-Voice accuracy.
Tesla and Starlink deploy the technology
The launch also highlights how voice AI is moving into customer-facing operations at major U.S. companies connected to xAI. The company said Grok Voice 3.0 scored 74% on ServiceNow’s EVA-Bench Accuracy test for enterprise voice agents.
Tesla reportedly uses the model to handle millions of conversations each day. Starlink uses it for about 20,000 daily customer calls, resolving 70% without human intervention.
Comments
Post a Comment