xAI released Grok Voice Think Fast 2.0, a speech-to-speech model for developers building voice agents, at $0.08 per minute of audio. The company reports faster time to first audio, better transcription in 24 languages, and more reliable tool calls, with no prompt changes needed. On 5 August 2026 the default voice alias switches to the new model, so anyone wanting the old one must pin it.
What changed
Voice agents ran on Think Fast 1.0, which was slower to start speaking and less accurate at transcription.
What it unlocks
Building voice agents that begin replying in under a second and transcribe more accurately over noisy phone lines, without rewriting existing prompts.
- $0.08 per minute of audio
- 82.9% vs 75.7% on speech benchmark
- first audio 1.25s to 0.70s
- alias switches 5 August 2026
What you need to act on it
- developer access to xAI's voice API
- pinning version 1.0 before 5 August 2026 to keep the older model
- testingcatalog.com2026-07-29