SpaceXAI has launched Grok Voice Transcribe 2.0, a speech-to-text API available under the model ID grok-voice-transcribe-2.0. The company has not released open weights, meaning self-hosting is currently not supported. The model is built on the foundation of the existing Grok Voice system, which currently processes tens of thousands of customer support calls daily and powers the Grok assistant in Tesla vehicles.
The new version features improved multilingual support and automatic language detection, including the ability to handle language switches during a single recording. On short phrases, such as in-car commands, the word error rate has decreased from 20.6% to 6.8%. The model supports 25 languages for formatting numbers, currencies, and units, and is currently ranked first among 32 streaming models on the Artificial Analysis leaderboard.
Pricing for the service remains unchanged from the previous version. Batch transcription is priced at $0.10 per hour, while streaming costs $0.20 per hour. These rates correspond to approximately $1.67 and $3.33 per 1,000 minutes of audio, respectively. The API includes features such as diarization, timestamps, and key term extraction, supporting files up to 500 MB in 12 different audio formats.
Atlassian Loom has officially adopted Grok Voice Transcribe 2.0 for its video transcription services. The integration allows users to record action plans in Loom, which are then transcribed and processed into code updates via the Cursor platform. This workflow is intended to streamline the transition from project context to code execution.
