AI News (2026/7/31): xAI Launches Grok Voice Think Fast 2.0, End-to-End Voice Model Responds in Just 0.70 Seconds

2026年7月31日 06:19

Executive Summary:

Product Overview

xAI has released Grok Voice Think Fast 2.0, a next-generation voice model built on an end-to-end speech-to-speech architecture. It directly processes spoken input and generates spoken output, eliminating the need for separate speech recognition, language model inference, and speech synthesis stages. The model achieves a first audio response time of 0.70 seconds, down from 1.25 seconds in the previous version, according to xAI. The company has not disclosed the model’s parameter count or the full list of supported languages.

Core Features and Performance Metrics

Grok Voice Think Fast 2.0 performs well on several benchmarks. In the Artificial Analysis speech-to-speech evaluation, it achieved a quality index of 82.9%, compared to 79.1% for OpenAI’s GPT-Realtime-2.1 High and 69.5% for Google’s Gemini 3.1 Flash High. In xAI’s internal agent test, the model scored 56.5%, versus 45.7% for GPT-Realtime-2.1 High.

For transcription, xAI reports that the model supports 24 languages and improves speech recognition accuracy by 1.5 to 2 times over professional transcription models such as Deepgram Nova 3 and ElevenLabs Scribe v2. The model also supports full-duplex conversation, allowing it to process speech while the user is still talking, without waiting for the input to finish.

Pricing and Usage

Grok Voice Think Fast 2.0 uses a unified pricing model: $0.08 per minute of audio, covering both input and output. xAI says the model has been tested in Starlink customer service scenarios, leading to improved sales conversion and automated resolution rates. For usage details, see the official release page.