xAI releases Grok Voice 3.0
xAI has released Grok Voice 3.0, a speech-to-speech model that ranks first on Artificial Analysis's independent index and is offered in the API at the same price as the previous version.
xAI has released Grok Voice 3.0, which it calls "the world's most intelligent speech-to-speech model." It is now available in xAI's Voice Agent API at $0.08 per minute ($4.80 per hour), the same price as Grok Voice 2.0. Agents that use the grok-voice-latest alias have already been switched to the new model with no code changes.
In independent testing by Artificial Analysis, the model scores 86.9 on the Speech-to-Speech Index and ranks first. The other scores were 83.2 for GPT-Live-1 (Astra, Medium), 82.6 for Gemini 3.8 Live (Ext. Thinking, High) and 81.3 for Grok Voice 2.0. The index combines speech reasoning, τ-Voice and the results of blind head-to-head calls. Other reported results:
- Big Bench Audio (speech reasoning): 98.5%
- Speech Agent Arena Elo: 1070, compared with 1048 for GPT-Live-1
- Task success rate: 94.8%
- Full Duplex Bench (handling pauses, turn-taking and interruptions): 97.0%
- τ-Voice (support calls with varied accents, background noise and interruptions): 71.6%. xAI says this is comparable to GPT-Live-1, with 40% lower latency.
- ServiceNow's EVA-Bench: 74% accuracy, the highest of the models tested
xAI says response latency is under one second. According to Artificial Analysis's measurements, it costs less per task than GPT-Live-1 and GPT-Realtime-2.1.
xAI also reported how the model is used in production. It handles millions of conversations a day in Tesla vehicles and powers Grok Bot's voice mode. Starlink call centers answer 20,000 calls a day with it in multiple languages. xAI says about 70% of those calls are resolved without a human, and one in four sales calls ends in a purchase. xAI did not publish independent figures to verify these numbers.