xAI launched Grok Voice Think Fast 2.0 on July 29, positioning it as its most capable speech-to-speech voice model to date. The model scored 82.9% on Artificial Analysis's overall speech-to-speech quality index, outperforming GPT-Realtime-2.1 High (79.1%) and Gemini 3.1 Flash High (69.5%).
Key improvements include a time-to-first-audio of just 0.70 seconds (down from 1.25s in version 1.0), a 95.1% score on conversational dynamics (up from 77.8%), and a 56.5% agentic performance rating versus 45.7% for GPT-Realtime. On transcription accuracy, xAI claims a 1.5–2x improvement over Deepgram Nova 3 and ElevenLabs Scribe v2, widening to ~10x in noisy environments.
The model uses a unique approach where it reasons through queries while speaking, enabling tool calls to execute before the first sentence finishes. On Starlink customer service (+1 888 GO STARLINK), A/B testing showed significant increases in both sales conversion rate and support containment rate.
Grok Voice Think Fast 2.0 is priced at $0.08 per minute of audio and will become the default voice model on August 5 when the grok-voice-latest alias upgrades from 1.0 to 2.0.
