Use Grok Voice Transcribe 2.0 on the Speech-to-Text API

use-grok-voice-transcribe-2-0Updated 2026-09-22

grok-voice-transcribe-2.0 is the current Speech-to-Text model on the xAI API. Docs list it as the default when model is omitted. News: Introducing Grok Voice Transcribe 2.0 (Sep 18, 2026). Capability guide: Speech to Text.

Unary call with the model named

curl -X POST https://api.x.ai/v1/stt \
  -H "Authorization: Bearer $XAI_API_KEY" \
  -F model=grok-voice-transcribe-2.0 \
  -F format=true \
  -F language=en \
  -F "keyterm=Understand The Universe" \
  -F file=@audio.mp3

Put every option field before file. The guide warns that fields after file may be ignored on streamable uploads.

Omit model and you still get grok-voice-transcribe-2.0. Naming it makes the choice visible in logs and code review.

Streaming

Open wss://api.x.ai/v1/stt with model=grok-voice-transcribe-2.0 in the query string (same default if omitted). Wait for transcript.created, send raw audio as binary frames, finish with {"type":"audio.done"}. Proxy the socket through your backend so the API key never sits in the browser.

What ships with 2.0 at the same price

Batch is $0.10 / hr of audio; streaming is $0.20 / hr (pricing). Diarization, timestamps, and key terms are included. Existing STT integrations that omit model pick up 2.0 accuracy with no code change once it is the default.

Related knobs

Multichannel, diarize, Smart Turn, filler words, and key terms: Transcribe audio with the Grok Voice API and Stream speech to text over WebSocket.
Stay on 1.0 during the transition: Pin Grok Voice Transcribe 1.0.

Pitfalls

Max file size is 500 MB (413 when larger).
format=true needs language.
This covers the Voice API STT endpoints only. grok.com Voice Mode and Grok Bot dictate / Start voice chat use different surfaces.