Overview
Grok Voice Transcribe 2.0 is the speech-to-text model xAI announced on 18 September 2026, served through the same Speech-to-Text API as its predecessor. xAI says it "transcribes dozens of languages, detects the language automatically, and follows mid-recording switches in a single pass", and the xAI voice documentation lists 25 languages and 12 audio formats for the transcription models.
The headline claim is accuracy at an unchanged price: xAI says the model is twice as accurate as Grok Voice Transcribe 1.0 at the same price, and that it ranks first for accuracy among 32 streaming models on the public Artificial Analysis leaderboard. On multilingual short phrases, xAI reports word error rate dropping from 20.6% to 6.8%. On an internal telephony evaluation over 8 kHz English customer-support audio, it says the model leads every competitor it tested.
Both transcription shapes are available: a single call for recorded files and URLs, and a WebSocket stream for real-time audio. Speaker diarization labels each speaker "at no additional cost", every word carries start and end times plus a confidence score, and the docs also list Smart Turn end-of-turn detection for voice-agent use.
Pricing is unchanged from 1.0: xAI lists batch transcription at $0.10 per hour of audio and streaming at $0.20 per hour, with diarization, timestamps and key terms included. Callers select the model explicitly as `grok-voice-transcribe-2.0`; xAI's release notes say `grok-voice-transcribe-1.0` remains the default and that 2.0 "will soon be the default in the Speech-to-Text API", so existing integrations pick up the improvement with no code changes once it flips.
| Released | 2026-09-18 |
|---|---|
| License | Proprietary |
| Weights | API only |
| Parameters | Not disclosed |
| Modalities | Audio, Text |
| Status | Generally available |
Pricing
| Input | $0.10 / hour of audio |
|---|
Batch transcription $0.10 per hour of audio; streaming $0.20 per hour. xAI says diarization, timestamps and key terms are included, and that the price is unchanged from Grok Voice Transcribe 1.0.
Strengths
- xAI reports twice the accuracy of Grok Voice Transcribe 1.0 at the same price
- Ranks first for accuracy among 32 streaming models on the public Artificial Analysis leaderboard, per xAI
- Multilingual short-phrase word error rate falls from 20.6% to 6.8% against 1.0, by xAI's measurement
- Speaker diarization included at no additional cost, with per-word start/end times and confidence scores
- Automatic language detection that follows mid-recording language switches in a single pass
- Batch transcription of files and URLs plus real-time streaming over WebSocket, from one API
Best for
- Reach for it for call-centre and telephony transcription, the case xAI benchmarked on 8 kHz English support audio
- Reach for it for real-time voice agents, where the WebSocket stream and Smart Turn end-of-turn detection apply
- Reach for it for meeting and interview transcripts that need speaker labels and word-level timestamps
- Reach for it for multilingual recordings where the language is unknown up front or changes mid-recording
How to access
| Provider | Model ID |
|---|---|
| xAI Speech-to-Text API ↗ | grok-voice-transcribe-2.0 |
FAQ
What is Grok Voice Transcribe 2.0?
It is the speech-to-text model xAI announced on 18 September 2026, served through xAI's Speech-to-Text API as `grok-voice-transcribe-2.0`. It handles both batch transcription of recorded files and URLs and real-time streaming over WebSocket.
How accurate is Grok Voice Transcribe 2.0?
xAI says it is twice as accurate as Grok Voice Transcribe 1.0 at the same price, and that it ranks first for accuracy among 32 streaming models on the public Artificial Analysis leaderboard. On multilingual short phrases xAI reports word error rate falling from 20.6% to 6.8%.
What does Grok Voice Transcribe 2.0 cost?
xAI lists batch transcription at $0.10 per hour of audio and streaming at $0.20 per hour, with speaker diarization, timestamps and key terms included. That is the same price as version 1.0.
How many languages does it support?
xAI's voice documentation lists 25 languages and 12 audio formats for its transcription models. The announcement says the model detects the language automatically and follows mid-recording language switches in a single pass.
Is it the default model in the API?
Not at announcement. xAI's release notes say both `grok-voice-transcribe-1.0` and `grok-voice-transcribe-2.0` are available with 1.0 as the default, and that 2.0 "will soon be the default in the Speech-to-Text API".
Does it label who is speaking?
Yes. xAI says speaker diarization labels each speaker in the transcript at no additional cost, and that every word carries precise start and end times plus a confidence score.