xAI releases Grok Voice Transcribe 2.0 for its speech-to-text API

xAI released Grok Voice Transcribe 2.0 for batch and streaming speech-to-text. Developers can choose the new model through the company’s existing API, and xAI said it improves transcription accuracy without changing speech-to-text prices.

For batch jobs, developers select grok-voice-transcribe-2.0 through the model parameter on the existing /v1/stt REST endpoint. Streaming WebSocket sessions use a query parameter, according to xAI’s speech-to-text documentation. xAI said existing integrations can use version 2.0 without code changes, although that compatibility claim has not been independently tested.

The release supports file- or URL-based batch transcription and real-time streaming. Documented controls include word-level timestamps, optional speaker diarization, independent transcription for up to eight audio channels, as many as 100 key terms, text formatting, filler-word handling, voice-activity thresholds and Smart Turn end-of-turn detection.

The model documentation lists version 2.0 at the same rates as version 1.0: $0.10 per audio hour for batch transcription and $0.20 per audio hour for streaming. Diarization, timestamps and key terms are included. For market context, Microsoft’s MAI-Transcribe-2 preview also entered the speech-to-text API market in September.

xAI said its real-world evaluations found version 2.0 twice as accurate as version 1.0 and better across four internal production-traffic test sets. It also said word error rate on an internal short-phrase test covering 19 languages fell from 20.6% to 6.8%. These are vendor claims based partly on unpublished internal evaluations.

xAI also claimed version 2.0 ranked first for accuracy among 32 streaming models on Artificial Analysis. The Artificial Analysis leaderboard describes an index built from three weighted word-error-rate datasets, but its accessible text did not expose the model-level ranking needed to confirm xAI’s placement or comparison count. The transcription launch follows its Grok 4.6 model update in August.

xAI’s materials do not agree on the transition status. The September 17 release notes identify version 1.0 as the default, while the live speech-to-text guide and model page observed on September 19 identify version 2.0 as the default when the model is omitted. xAI said version 2.0 would become the default and version 1.0 would be deprecated in the coming weeks. Developers who still require version 1.0 can pin grok-voice-transcribe-1.0, but xAI did not provide a deprecation date.