Grok Voice 3.0 leads voice AI rankings

SpaceXAI has introduced Grok Voice 3.0, a new AI model for real-time voice conversations. It is designed primarily to handle complex customer service requests more reliably on its own, respond more naturally to pauses and interruptions, and begin speaking in less than one second on average.

In independent tests by Artificial Analysis (AA), Grok Voice 3.0 leads the voice models tested. It scores 86.9 points in the Speech-to-Speech Index. It is followed by OpenAI’s GPT-Live-1 at 83.2 and Google’s Gemini 3.8 Live at 82.6 points. Its predecessor, Grok Voice 2.0, scored 81.3 points. The overall score factors in logical reasoning on spoken tasks, the ability to complete tasks autonomously, human testers' preferences, and the success rate for completing tasks in voice conversations.

Leading the overall index doesn’t mean it ranks first in every category. In blind tests, where people talk with the voice models and choose their favorite, Grok Voice 3.0 ranks third behind two Gemini models. It leads in task completion with a score of 94.8%, though only narrowly ahead of its predecessor.

Millions of voice conversations daily at Tesla

The biggest improvement is on the τ-Voice benchmark. The test simulates customer service conversations involving different issues and accents and checks whether the AI successfully completes the tasks. The success rate rises from 56.5% with the predecessor to 71.6%. That puts Grok Voice 3.0 in second place, behind GPT-Live-1 (Astra, Medium) at 74.5%.

Response speed also remains high: AA measures an average of 0.81 seconds before speech output begins. The top configurations from OpenAI and Google that were compared take about 1.3 seconds. Grok Voice 2.0 was slightly faster at 0.7 seconds.

The price remains unchanged at $0.08 per minute. In Artificial Analysis' cost comparison, Grok Voice 3.0 sits in the middle of the five highest-ranked voice models at $4.80 per hour of input audio.

According to SpaceXAI, Grok Voice 3.0 already handles millions of conversations daily in Tesla vehicles and is used in Grok Bot’s voice mode. At Starlink, Grok Voice handles around 20,000 customer calls on weekdays, about 70% of which are completed without human assistance. The voice agent is supposed to resolve technical issues, issue credits, and arrange replacement shipments.

Grok Voice 3.0 is available now through SpaceXSI’s API.

(mho)