x.ai released Grok Voice Transcribe 2.0. It is twice as accurate as its predecessor at the same price. It ranks first for accuracy among 32 streaming models on Artificial Analysis. It handles noisy, multilingual, and real-world audio better than prior versions. It supports batch and streaming transcription with word-level timestamps. Speaker diarization, multichannel transcription, and key term biasing are included. Text formatting and filler word removal are built in. It will soon replace Grok Voice Transcribe 1.0 as the default in the Speech-to-Text API. Grok Voice Transcribe 1.0 will be deprecated in the coming weeks. Atlassian Loom adopted it for higher accuracy in video transcription.
What changed
Grok Voice Transcribe 1.0 was the default speech-to-text model.
What it unlocks
Transcribing noisy, multilingual, or real-world audio with higher accuracy at the same cost.
- 2x accuracy vs Grok Voice Transcribe 1.0
- $0.10 per hour batch
- $0.20 per hour streaming
What you need to act on it
- use the Speech-to-Text API
- pin `grok-voice-transcribe-1.0` to delay migration
Sources