Batch rank still matters for files
A streaming trophy does not change the cost or timing contract of a three-hour upload. MAI remains the file default while that batch price and output shape hold.
BATCH VS LIVE
Grok Voice Transcribe 2.0 vs MAI-Transcribe-2
Use Grok for the live room and MAI for uploaded files. The two models are not interchangeable just because both transcribe speech.
Pricing and leaderboard ranks checked September 20, 2026
Different jobs
Streaming accuracy does not automatically win a subtitle pass.
Grok Voice Transcribe 2.0 leads Artificial Analysis streaming accuracy. On the separate batch board it placed fifth at 2.3% WER, behind MAI-Transcribe-2 at 2.0%. CastTranscript keeps those contracts apart: one streaming pass for the room, one batch pass for files and optional refinement.
Decision pointGrok Voice Transcribe 2.0MAI-Transcribe-2
CastTranscript role
Default live recording path
Default upload path and explicit live refine pass
Leaderboard fit
First on AA-WER Streaming final and first-partial accuracy
2.0% batch WER on the Artificial Analysis results cited with the 2.0 launch
Published price
$0.20 per hour streaming, $0.10 per hour batch
$0.10 per audio hour as a limited offer through the end of 2026
Output contract
Interim text, word timing, diarization and turn detection for a live document
Long-form transcription with word timestamps and diarization for files and subtitles
When to choose it
Meetings, interviews and any session that must update while people are speaking
Podcasts, finished video, MP3 archives and a publishable subtitle pass
Mixing them
Not used to backfill a MAI file job
Not used to silently rewrite a live Grok transcript unless the user starts Refine
Live Grok does not become fake SRT
Word times from the live pass stay attached to that pass. CastTranscript will not invent a subtitle track by merging Grok turns with a second model’s cues.
Refine is a second paid contract
After a live session, Refine with MAI is optional. It is another credit per media minute on the same private document, with the model name recorded on the job.
CastTranscript decision
Grok for the room. MAI for the file.
Live audio uses Grok Voice Transcribe 2.0. Uploads and explicit refinement use MAI-Transcribe-2. The editor stays the same so the model can change without changing the export workflow.
Try an uploadQuestions, answered
Grok and MAI routing questions
Why doesn’t CastTranscript use Grok for uploaded files?
Grok leads the streaming accuracy board, but the current batch ranking and the existing MAI file contract still fit long podcasts and subtitle work. Uploads stay on MAI-Transcribe-2.
Can I refine a live Grok transcript with MAI?
Yes. Refine is an explicit second pass on the same document. It is a separate per-minute charge and does not run automatically.
Does CastTranscript combine Grok text with MAI timestamps?
No. Each pass owns its timing and speaker fields. Missing detail is disclosed instead of being stitched from another model.
Is the MAI $0.10 per hour price permanent?
No permanent price is assumed. Microsoft described $0.10 per audio hour as a limited offer through the end of 2026, so this page includes its verification date.
Private by default
Use the pass that matches the source.
Upload a file for MAI, or record live with Grok, then keep one private editor either way.