SpaceXAI on September 18, 2026 released Grok Voice Transcribe 2.0. The company's official press release said the model is built on the audio foundation model that powers Grok Voice, and described it as one of the most accurate transcription models available in real-world company evaluations, claiming it is "twice as accurate" as 1.0 (the company's own claim), while pricing remains unchanged.
Public pricing is unchanged from 1.0: US$0.10 per audio hour for batch transcription and US$0.20 per hour for streaming, with features including speaker diarization, timestamps and keyterm prompting. The company said 2.0 will soon become the default for the Speech-to-Text API; 1.0 will be phased out over the coming weeks, and grok-voice-transcribe-1.0 can be pinned during the transition.
SpaceXAI said Grok Voice already processes tens of thousands of customer service calls daily, transcribes millions of hours of video narration, and runs on physical products including Tesla's in-car Grok. The company also said that on Artificial Analysis's public streaming leaderboard, Transcribe 2.0 ranks first in accuracy among 32 streaming models — this is the company's characterization of the leaderboard.
The official release also mentioned Atlassian Loom: Atlassian said 2.0 is more accurate than its previous solution and is being used for transcribing Loom videos. Sanchan Saxena, senior vice president of Teamwork Collection, said that pairing Grok-powered Loom voice-to-text with Cursor represents a shift in which "context flows into actual work."
Product details are available in the SpaceXAI press release: https://x.ai/news/grok-voice-transcribe-2.