Grok Voice Think Fast 2.0 Debuts With Higher Accuracy, Faster Responses, and Lower Latency

SpaceXAI today introduced Grok Voice Think Fast 2.0, a new speech-to-speech AI model that the company says delivers higher intelligence, improved transcription accuracy, faster responses, and more natural conversations. The release succeeds Grok Voice Think Fast 1.0 and will become the default version behind the grok-voice-latest endpoint next month.

According to SpaceXAI, Grok Voice Think Fast 2.0 improves performance across reasoning, conversational ability, and tool use while reducing response latency. The company cited benchmarking from Artificial Analysis showing the model achieved an overall Speech-to-Speech Quality Index score of 82.9%, compared with 75.7% for Grok Voice Think Fast 1.0, 79.1% for GPT-Realtime-2.1 (High), and 69.5% for Gemini 3.1 Flash (High).

The company also reported gains in several benchmark categories, including speech reasoning, conversational dynamics, and agentic performance. It said the model produces its first audio response in 0.70 seconds, compared with 1.25 seconds for the previous version.

A major focus of the update is transcription accuracy. SpaceXAI said Grok Voice Think Fast 2.0 was evaluated on thousands of short phrases spanning 24 languages and achieved between 1.5x and 2.0x better accuracy than Deepgram Nova 3 and ElevenLabs Scribe v2. The company also reported a 1.4x improvement over Grok Voice Think Fast 1.0.

The performance gap widened further in noisy environments, according to SpaceXAI. The company said the new model delivers roughly 10 times better accuracy than dedicated speech-to-text systems under conditions that include background noise and telephony compression.

SpaceXAI also redesigned the model's reasoning process to improve efficiency. Unlike many speech models, Grok Voice Think Fast performs reasoning while speaking rather than before generating audio. The company said Grok Voice Think Fast 2.0 requires approximately 40% of the reasoning tokens used by its predecessor, allowing tool calls to complete more quickly, often before the model finishes its opening sentence.

The release also emphasizes more natural dialogue. SpaceXAI said it used reinforcement learning to train the model to produce shorter responses, ask one question at a time, and reduce unnecessary conversational filler. The goal is to create smoother interactions while the model continues handling more complex reasoning in the background.

The company said existing applications should see improved performance without requiring prompt changes. During A/B testing on Starlink's customer support line, SpaceXAI said the model increased both sales conversion rates and support containment rates.

SpaceXAI plans to migrate the grok-voice-latest endpoint from Grok Voice Think Fast 1.0 to Grok Voice Think Fast 2.0 on August 5, 2026. Developers who want to continue using the previous version will need to pin grok-voice-think-fast-1.0 before that date.

Grok Voice Think Fast 2.0 is priced at $0.08 per minute of audio.

This analysis is based on reporting from x.ai.

Image courtesy of SpaceXAI.

This article was generated with AI assistance and reviewed for accuracy and quality.