xAI has released Grok Voice Think Fast 2.0, a voice agent model designed for customer service, sales, and phone scenarios. It can understand users, generate spoken responses, and call tools. The new version focuses on improved reasoning, transcription, and conversation abilities.
Independent testing by Artificial Analysis gives it an overall voice index of 82.9%, ranking second behind Alibaba's Qwen Audio 3.0 Realtime Plus at 84.1%. But on the τ-Voice benchmark, which tests voice agents on customer service tasks, it scored 56.5% — first place, ahead of Qwen's 54.6% and GPT-Realtime-2.1 High's 45.7%. Time to first audio is just 0.70 seconds.
That performance comes with a price hike: the API cost has gone from $0.05 per minute to $0.08 per minute, a 60% increase. That works out to $4.80 per hour of audio.