xAI announced Grok Voice Transcribe 2.0, a new version of its speech-to-text transcription model.
The company targets uses where audio is recorded in real conditions, rather than only clean studio recordings.
What xAI announced
According to xAI, the new model is based on the same voice infrastructure used by Grok Voice.
The company says the system was trained on multilingual audio recorded under varied conditions, including noisy environments.
xAI claims the new model is twice as accurate as the previous version in its internal evaluations, with no price change.
It is important to emphasize that these are claims and measurements published by the company itself.
Why accurate transcription matters
Transcription has become important infrastructure in many AI systems.
It is used in voice agents, service centers, meeting summaries, subtitle creation, video search and more.
In these uses, small speech recognition errors can affect every subsequent stage.
What this means for voice product builders
Developers and companies building voice-based products can add the model to the options they test alongside other transcription services.
As with any such service, test it on the product's real audio rather than relying solely on the provider's benchmark figures.
Hebrew
xAI describes the model as multilingual, but AI24 has not verified Hebrew transcription quality at this stage.
General performance does not mean Hebrew quality matches other languages.