There isn’t any public indication of a separate “seeded prompt inference” launch date for Gemini 3.1 Flash TTS; the model is already in preview and was introduced in April 2026. Google’s announcement says it started rolling out for developers, enterprises, and Workspace users “starting today,” and the API docs also mark the latest update as April 2026.

What is known

  • Gemini 3.1 Flash TTS Preview is publicly available now in preview form.
  • Google’s prompt guide describes the model as natively interpreting transcripts and using prompts/tags to control delivery.
  • Public release notes and articles around launch do not mention a planned feature called “seeded prompt inference.”

Practical read

If by “seeded prompt inference” you mean a way to make voice/style generation more reproducible, that capability does not appear to have a published ETA yet. The safest expectation is that any such feature would come through a future API update or changelog entry rather than a separate announcement.

Bottom line

The current answer is: no announced date. It’s already out in preview, but there’s no public confirmation of when or whether seeded prompt inference will be added.