Skip to main content

Why is transcription sometimes fast and sometimes slow?

The GPU hosting the model starts up on demand: the first transcription after a quiet period can take a while (a cold start), after which subsequent pages run much faster — so batching is quickest. Occasionally jobs queue behind another user's bulk run, and pages where Leo detects a poor result and automatically retries take several times longer than a clean pass. If a page has shown no progress for a very long time, message us.

Did this answer your question?