xAI Releases Grok Voice Think Fast 2.0 Voice Agent Model
xAI (SpaceXAI) has released Grok Voice Think Fast 2.0, a voice agent model designed for customer service, sales, and telephone scenarios, capable of understanding users, generating voice responses, and calling tools.
The new version focuses on enhancing reasoning, transcription, and dialogue capabilities. An independent evaluation by Artificial Analysis shows a comprehensive voice index of 82.9%, ranking second, and a τ-Voice customer service task benchmark of 56.5%, ranking first, with the first response taking only 0.70 seconds.
The event-driven voice agent market is accelerating competition, with funding flowing towards models with low latency and high tool-calling capabilities. xAI benefits from its advantages in customer service and sales applications, with prices rising from $0.05 to $0.08 per minute, putting pressure on competitors.
Source: Public Information
ABAB AI Insight
xAI previously launched the Grok Voice series and applied it in its own app and collaborative scenarios. Now, with Think Fast 2.0, it strengthens full-duplex dialogue and tool reliability, targeting high-value telephone and customer service workflows rather than general chatting.
On the capital path, the company links model iterations with API pricing, motivated by exchanging higher performance for premium pricing, while validating advantages with independent benchmarks, focusing resources on measurable improvements in reasoning and task completion rates.
Similar to the early transition of voice assistants from toys to enterprise tools, or the OpenAI Realtime API promoting real-time voice agents, the current phase is shifting from latency optimization to agentic task completion in voice models.
Essentially, this represents a technological substitution: voice agents are replacing some human customer service and sales calls, as sub-second responses and reliable tool calling make end-to-end task automation possible, thereby restructuring the cost structure of enterprise services.
ABAB News · Cognitive Laws
The true threshold for voice agents is task completion rate.
Low latency combined with tool calling is the production-level capability.
Price increases often accompany verifiable performance leaps.