xAI’s New Voice Model Just Beat OpenAI and Google

xAI put out a new version of its voice assistant model, Grok Voice Think Fast 2.0, and the jump from version one is bigger than most model updates get.

On the independent Artificial Analysis benchmark, it scored 82.9 percent on the Speech to Speech Index, up from 75.7 for the first version, and ahead of OpenAI’s GPT Realtime 2.1 at 79.1 and Google’s Gemini 3.1 Flash at 69.5. On a separate test for how well a voice model can act as an agent and actually get things done, it topped the list at 56.5 percent, ahead of the first version’s 52.1 and every other model on the board.

The speed numbers are the part that’ll actually change how it feels to use day to day. Time to first audio dropped from 1.25 seconds to 0.70, which is the difference between a voice assistant that feels like it’s thinking and one that feels like it’s just talking to you. Transcription accuracy improved across two dozen languages, with the biggest gains showing up in noisy conditions like a phone call on a busy street or a shaky connection.

It’s priced at 8 cents a minute, reasoning token use is down around 60 percent compared to the old version, and everyone still on the older Think Fast 1.0 gets moved over automatically on August 5, no action needed on the user’s end.

Voice is the interface that’s easy to take for granted until it’s bad, the lag, the mishearing, the flat robotic cadence that makes you switch back to typing. Shaving half a second off a response doesn’t sound like news, but it’s the difference between an assistant you talk over and one you actually talk to. That race just got more competitive, and the prices are still heading down, not up.

——

Follow: @Ali Demi
Book your free AI clarity call, NOW!
https://buff.ly/TpWy277

——

Sources:
https://www.testingcatalog.com/spacexai-launches-grok-voice-think-fast-2-0-on-agent-builder/
https://finance.biggo.com/news/4bf16f12-1046-46f1-8d01-aa9649230336
https://x.com/ArtificialAnlys/status/2082528987272957960

Repost this. Thanks.