Grok Voice Think Fast 1.0
xAI · United States · 2026
xAI's first speech-to-speech model — superseded as the default in August 2026, but still available and still the cheaper option.
Grok Voice Think Fast 1.0 was released on 23 April 2026 as xAI's first realtime voice agent model, streaming audio in both directions over a WebSocket connection rather than assembling a conversation from separate transcription, language and synthesis services. It held the default position for just over three months. On 29 July 2026 xAI announced version 2.0, and on 5 August the `grok-voice-latest` alias moved across, so any application that had not pinned a specific model identifier was upgraded without doing anything. Version 1.0 was not withdrawn: developers can still call `grok-voice-think-fast-1.0` explicitly, and the reason to do so is money. It costs $0.05 per minute of audio against $0.08 for the newer model — a 37% saving for workloads where quality is already sufficient. The gap between the generations is documented. On the Artificial Analysis speech-to-speech quality index version 1.0 scores 75.7 against 82.9 for its successor, and xAI reports the newer model achieving 1.4 times the transcription accuracy across thousands of short phrases in 24 languages. No end-of-service date has been announced for version 1.0.
▸Videos
No videos yet.
▸Reviews
No reviews yet. Be the first!