Grok Transcribe 1.0 now routes to 2.0 as old model retires
Grok Voice Transcribe 1.0 requests now run on version 2.0. See what changed on October 2, current API prices and the checks developers should make.
SpaceXAI has retired Grok Voice Transcribe 1.0. Its October 2 release note says every request using the older model identifier now goes to Grok Voice Transcribe 2.0 at the same price. Keeping the old name in a configuration no longer keeps the old model.
That changes the transition advice in the company’s September 18 announcement. At launch, developers could explicitly select version 1.0 to remain on it temporarily. The latest notice closes that option. The relevant news is the completed retirement, rather than another launch of version 2.0.
What the old model setting does now
The current implementation guide documents the same routing for both recorded audio and live streams. A request without a model setting uses version 2.0. A request explicitly naming version 1.0 also reaches version 2.0.
Model selection remains a field in the REST form for uploaded audio and a query parameter for WebSocket streams. The guide continues to document transcripts, audio duration and word timestamps, with speaker or channel information when the relevant options are enabled.
For teams that used the older identifier as a stability control, this is the important boundary. A successful request can still run on a different model from the one a configuration appears to select. Update deployment records so they reflect what the service now delivers.
Pricing and service limits
The current model specification lists recorded audio transcription at $0.10 per audio hour and streaming at $0.20 per hour. It lists ten requests per second for each route and 100 concurrent streaming sessions per team. These are current service details, not new October 2 features.
The implementation guide also retains a 500 MB file limit and support for up to eight audio channels. Streaming with the Opus audio format is limited to a single channel. Filler words are removed by default, and formatting spoken numbers into written values depends on the selected formatting options.
Check the transcript before relying on it
ByteForward recommends rerunning representative recordings and any downstream checks after this routing change. Include product names, numbers, overlapping speakers and the actual audio conditions your application encounters. Compare the resulting text and timestamps with what the rest of the system expects.
SpaceXAI reported accuracy improvements in its September launch evaluations. Those vendor results provide background, but they do not guarantee identical text or timing for a particular recording. Existing integrations continuing to work does not remove the need to check the output they consume.
For related developments, ByteForward’s Microsoft streaming transcription report looks at another API option. Here, the immediate task is simpler. Treat Grok version 1.0 as retired and test the replacement already serving those requests.
Featured photo by Anna Pou on Pexels, used under the Pexels License. Illustrative recording equipment.
