Grok’s older transcription model has reached its retirement date. According to the official xAI release notes, grok-voice-transcribe-1.0 reached end of life on October 2, 2026. Requests using that model name are now routed to grok-voice-transcribe-2.0 at the same price. The provider describes the replacement as more accurate, without publishing a quantified improvement in this retirement notice.
For publishers turning interviews, podcasts and video into searchable content, the practical consequence is easy to miss: an unchanged configuration can now produce transcripts using a different model. The old identifier no longer preserves the old transcription engine.
Automatic routing keeps requests moving
This is a retirement with a redirect rather than an announcement that every request using the previous name will fail. Version 2.0 was introduced in the September 17 release notes. The October notice confirms that requests to the retired slug are served by its successor.
Our operational recommendation is to update active configurations to the explicit 2.0 identifier and record the change in deployment notes. This makes the intended dependency clearer for editors, developers and future maintainers. It is housekeeping, rather than a requirement stated in the retirement announcement.
Request continuity should not be confused with identical output. A successful response tells a team that its integration still works; it does not establish that punctuation, proper names or individual words remain unchanged. Treat the transition as a reason to inspect representative results before allowing unattended publication.
Test the audio that matters to your business
A useful regression set starts with recordings your organisation actually publishes: an interview containing product names, a discussion with regional accents, a noisy recording, and a passage containing prices or dates. Compare the transcript against the original audio and a human-reviewed reference. Look especially for missing negations, confused speakers and entity names that could change the meaning of a quotation.
The Speech to Text documentation identifies grok-voice-transcribe-2.0 as the default model when the model parameter is omitted. It also documents word timestamps, key-term hints, speaker diarization and formatting options. These settings deserve review alongside the model identifier because editorial systems may depend on the structure and presentation of the returned text.
For example, a team generating subtitles should check timing alignment as well as wording. A team extracting quotations should listen to the relevant passage before publishing. A workflow producing summaries should verify whether an upstream transcription change affects the summary’s interpretation. These are proposed checks, not defects reported by xAI.
Keep the raw recording and the reviewed transcript together where appropriate. Record which configuration produced a new transcript, and avoid silently replacing an approved quotation with newly generated wording. Reprocessing an archive is a separate editorial decision; the routing change does not itself rewrite transcripts already stored in a CMS.
What this means for SEO and AI Search
The SEO angle is the quality of the material entering the publishing pipeline. If a transcript becomes the basis for an article, incorrect names or numbers can spread into headings, summaries and internal links. Reviewing the source text before generating those derivatives helps prevent one recognition error from becoming several published errors.
For AI Search, teams should distinguish the availability of a text version from its reliability. Publishing a transcript can make an audio discussion accessible to readers who prefer text, but this model retirement provides no evidence of a ranking improvement or an increase in AI citations. Neither the provider’s accuracy claim nor the unchanged price demonstrates a visibility gain.
Start with an inventory of transcription jobs, integration settings and documentation that still mention 1.0. Then validate a small, representative audio set and review any downstream publishing steps. The redirect reduces the immediate migration burden, while the responsibility for accurate quotations and trustworthy content stays with the publisher.