News in Short
- SpaceXAI has launched the Grok Voice Think Fast 2.0 AI model.
- The new model improves speech reasoning, transcription and conversational abilities.
- Users will be migrated to the new model starting August 5.
- SpaceXAI claims it outperformed OpenAI and Google models in several benchmarks.
- The AI voice model is priced at $0.08 per minute of audio.
Elon Musk-led SpaceXAI has unveiled Grok Voice Think Fast 2.0, the latest version of its voice AI model designed to deliver faster, more natural conversations and improved speech understanding. The new model succeeds Grok Voice Think Fast 1.0, which debuted earlier this year, and brings upgrades across transcription accuracy, speech reasoning and AI-powered voice interactions.
According to the company, the new model also outperformed competing voice AI models from OpenAI and Google in several internal benchmark tests.
Grok users will be upgraded automatically
SpaceXAI said existing Grok users will be migrated to Grok Voice Think Fast 2.0 on August 5, with no action required from users.
Those who wish to continue using Think Fast 1.0 can do so by manually pinning the older model before the migration begins.
The company has priced the new voice model at $0.08 per minute of audio, making it available for developers and businesses building AI voice experiences.
Better transcription, reasoning and voice conversations
The biggest improvements in Grok Voice Think Fast 2.0 focus on how the AI understands and responds to spoken conversations.
SpaceXAI says the model offers better speech reasoning, more accurate speech-to-text transcription, improved conversational flow and more reliable use of external tools during voice interactions.
The company also revealed that internal testing on Starlink showed a significant improvement in customer support performance and sales conversion rates during A/B testing.
SpaceXAI claims better benchmark performance
SpaceXAI compared the new model against Grok Voice Think Fast 1.0, OpenAI’s GPT-Realtime-2.1 and Google’s Gemini 3.1 Flash.
According to the company, Grok Voice Think Fast 2.0 scored 82.9 percent on the AA Speech-to-Speech Quality Index, outperforming GPT-Realtime-2.1, Gemini 3.1 Flash and its predecessor.
The model also achieved 97.2 percent in the Big Bench Audio Speech Reasoning benchmark, narrowly edging past the competing models.
However, SpaceXAI acknowledged that GPT-Realtime-2.1 performed slightly better in the Full Duplex Bench Conversational Dynamics test, scoring 95.7 percent compared to Grok’s 95.1 percent.
Supports multiple languages
Beyond benchmark scores, SpaceXAI says Grok Voice Think Fast 2.0 demonstrated 1.5x to 2x better transcription accuracy than Deepgram Nova 3 and ElevenLabs Scribe v2 during internal testing.
The evaluation reportedly covered thousands of short audio samples across 24 languages, highlighting improvements in multilingual voice recognition.
As competition in AI voice assistants continues to intensify, the launch of Grok Voice Think Fast 2.0 underscores SpaceXAI’s ambition to compete with Google and OpenAI by offering faster, more capable conversational AI for developers and enterprises.
