Grok Unveils Voice Think Fast 2.0 at $0.08 a Minute as Accuracy Jumps 1.5-2.0x
Updated
Updated · xAI · Jul 29
Grok Unveils Voice Think Fast 2.0 at $0.08 a Minute as Accuracy Jumps 1.5-2.0x
3 articles · Updated · xAI · Jul 29
Summary
August 5 is the switchover date for grok-voice-latest to Grok Voice Think Fast 2.0, with existing users upgraded automatically unless they pin version 1.0.
The new speech-to-speech model posts an 82.9% overall quality score and 1.25-second time to first audio, while Grok says it improves transcription accuracy by 1.5-2.0x versus Deepgram Nova 3 and ElevenLabs Scribe v2.
24 languages and noisy, telephony-compressed settings were a focus of the release, where Grok says the gap versus dedicated speech-to-text models widens to about 10x.
Tool calls are designed to finish before the agent's first sentence ends, reflecting lower reasoning-token use and a push toward shorter, more natural conversational turns.
Starlink A/B tests showed higher sales conversion and support containment rates, and Grok priced the model at $0.08 per minute of audio.
If Grok Voice 2.0 aggressively strips away natural filler words, does its hyper-efficient dialogue risk sounding more robotic than human?
With the default API switch looming on August 5, will Grok's parallel reasoning seamlessly upgrade or unexpectedly break enterprise workflows?
Can xAI’s claim of outperforming dedicated transcription giants truly hold up in chaotic real-world environments beyond their own controlled benchmarks?