xAI Makes Grok Voice Transcribe 2.0 the Speech-to-Text Default
The September model release now sits behind unpinned API requests, with batch and streaming paths, production-oriented transcript controls and unchanged listed hourly prices.
Edited by Tyronne Panaino
xAI released Grok Voice Transcribe 2.0 on September 18 as a new speech-to-text model for recorded and live audio. The official announcement introduced the model and its transcription controls, while xAI's current API documentation now identifies version 2.0 as the default.
That change matters to developers whose integrations use the service default instead of naming a model version. The current Speech-to-Text documentation lists both Grok Voice Transcribe 1.0 and 2.0 as selectable, so teams can still pin a version while they compare transcripts, latency and downstream behavior.
The API default has moved to version 2.0
The release announcement initially described version 2.0 as a model that would soon become the default. The current documentation provides the later checkpoint: `grok-voice-transcribe-2.0` is now the default for the Speech-to-Text API. That is a practical migration event, not merely a model announcement, because unpinned requests now resolve to the newer version.
Version 1.0 remains named as an available option in the current documentation. Teams with automated transcript review, regulated workflows or tightly tested post-processing can therefore hold their existing model choice while evaluating version 2.0. The documentation does not publish a final retirement date for version 1.0, so users should not infer a fixed migration deadline from the current page.
Controls cover batch and live transcription
xAI documents two delivery paths: a REST interface for file-based batch transcription and a streaming interface for real-time, low-latency use. The current service page lists support for several audio formats, multiple languages, interim streaming results, domain-specific key-term prompting and machine-learning-based end-of-turn detection.
The announcement adds more detail about transcript output and input handling. It says version 2.0 can return word-level timestamps and confidence scores, label speakers, process as many as eight channels independently, bias recognition toward as many as 100 supplied domain terms and remove filler words. Those controls are relevant to call centres, media transcription and voice-agent systems, but each team still needs to test how they behave with its accents, noise, vocabulary and audio pipeline.
Published prices and region remain narrow facts
The current API page lists REST transcription at $0.10 per audio hour and streaming transcription at $0.20 per audio hour. It also identifies `us-east-1` as the service region. Those details establish xAI's published interface, price and regional scope at review time; they do not establish account-specific availability, total operating cost or performance under a customer's workload.
The most consequential unsupported claims are about accuracy. xAI presents public-leaderboard and internal-evaluation comparisons in its announcement, but this article does not reproduce the rankings or treat them as independently verified. The external leaderboard was not fetched as evidence for this release, and internal production tests remain vendor-controlled. The next useful checkpoint is independent evaluation across representative languages, noisy audio, speaker overlap and domain terminology.
Status
Confirmed. xAI's announcement establishes the September 18 model release, and its current developer documentation confirms the default, supported interfaces, listed pricing and region. Internal confidence is medium because the operational facts come from official sources while comparative accuracy and real-world reliability have not been independently validated here.
Sources
Update note: Last reviewed 2026-09-24. We will revise this post if xAI changes the default, version availability, pricing, regional scope or publishes independently testable performance evidence.
Sources
- xAI — Introducing Grok Voice Transcribe 2.0 — official
- xAI Developers — Speech to Text — official
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.