ToolsNews
Grok Voice 3.0: xAI's new voice agent model, same price
xAI released Grok Voice 3.0, a speech-to-speech model, in its Voice Agent API on October 14 at $0.08 per minute, the same price as its predecessor. xAI says latency is under one second; the benchmark and production figures are as reported by xAI.
By: Mate · · 2 min read
xAI introduced Grok Voice 3.0 as the world's "most intelligent" speech-to-speech model (announcement). For developers the essentials are short: it's in the Voice Agent API at $0.08 per minute ($4.80 per hour), and agents on the grok-voice-latest alias have already been switched over automatically.
That last point cuts both ways. It's convenient, but if an agent's behaviour suddenly shifts, a model swap underneath may be why. Pinning a version in production is safer where possible.
The numbers, and where they come from
xAI cites Artificial Analysis measurements from October 13: 86.9 on the Speech-to-Speech Index (versus 81.3 for Grok Voice 2.0, 83.2 for GPT-Live-1 and 82.6 for Gemini 3.8 Live), and 97% on Full Duplex Bench, which measures handling of pauses, turn-taking and interruptions.
On the τ-Voice support-call test it reports 71.6%, which xAI says is comparable to GPT-Live-1 at 40% lower latency. On ServiceNow's EVA-Bench for enterprise voice agents it reports 74% accuracy. These figures were selected and published by xAI; independent verification is worth waiting for.
In production: Tesla and Starlink
xAI says the model handles millions of conversations a day in Tesla vehicles. Starlink's call centres use it for 20,000 calls a day across multiple languages; xAI says about 70% are resolved without a human, and one in four sales calls ends in a purchase. These are also the vendor's own numbers.
Why it matters: for voice agents, latency and interruption handling often matter more than raw intelligence. People hang up on bots that are slow or talk over them. xAI is putting its emphasis on exactly those metrics.
Before you wire it in
Test it on your own call patterns, especially in non-English languages, since the announcement gives no per-language results. Compare the per-minute price against your current setup using real call durations.
