grok-voice-latest repoints to Think Fast 2.0 tomorrow — pin 1.0 before then, or your voice agent changes model

If you call xAI's speech API through the alias grok-voice-latest, the model underneath it changes tomorrow. xAI's release notes state that on August 5, 2026, grok-voice-latest moves from grok-voice-think-fast-1.0 to grok-voice-think-fast-2.0. Nothing breaks and no action is required to upgrade — but if you have tuned prompts, latency budgets or evals against 1.0, pin grok-voice-think-fast-1.0 explicitly before then.

The 2.0 model was announced on July 29, 2026 as xAI's most capable speech-to-speech model. Artificial Analysis carries both versions on its Speech to Speech Index: 2.0 scores 82.9% against 1.0's 75.7%, and time to first audio falls from 1.25 seconds to 0.70. That combination is rarer than it sounds — the smarter model is also the faster one, which xAI attributes to the model reasoning while it is already speaking rather than before it starts. The docs list it as text-and-audio in, text-and-audio out, served from us-east-1, eu-west-1 and us-saltlake-2.

We are not quoting a price. Several write-ups report $0.08 per minute of audio; xAI's own model page renders its pricing table empty at the time of writing, so we cannot confirm that figure and will not repeat it as fact.

Why this took five days to reach the wire: the radar hunts text models. This is the second launch in two weeks that went straight through a blind spot in the watchlist rather than a failure in the alerting — see the FLUX 3 item posted alongside this one. grok-voice-think-fast-2.0 is now a tracked entry.