Google's deprecations table is a dry spreadsheet, and that is exactly why it is reliable: one row per model, with its dates and the replacement Google points to. One of those rows has September 30, 2026 in it.
What the table says
The row, verbatim, on the official Gemini API deprecations page:
gemini-omni-flash-preview· June 30, 2026 · September 30, 2026 ·gemini-omni-1.1-flash
What can be read from that row without inventing anything: the model is gemini-omni-flash-preview, the final date on the row is September 30, 2026, and the replacement Google indicates is gemini-omni-1.1-flash.
What the row does not say, and therefore stays out of this article: whether request limits change, whether the request format is the same, and whether the response comes back identical. The table is a calendar, not a migration guide.
What changes if you have automation in production
A model with preview in its name carries a built-in warning. It exists to be tested and replaced — and the replacement has a date.
The problem is almost never the migration itself. It is that the model call tends to be buried inside a workflow node someone configured in March and nobody has opened since. The flow runs, delivers, nobody touches it. Then the date arrives and the automation stops answering on a day you are looking somewhere else.
If you built any audio, image or multimodal flow in the last few months, it is worth opening it and checking which model string is inside.
In practice
- Grep for
gemini-omni-flash-previeweverywhere an API call lives — workflows, scripts, environment variables, agent configs. - Swap in the string
gemini-omni-1.1-flashin a test environment first and compare the output before shipping. The table names the replacement; it does not guarantee an identical response. - Put September 30, 2026 on the calendar and make the switch with margin, not on the eve.
Source
Google AI for Developers — Gemini API deprecations, ai.google.dev/gemini-api/docs/deprecations.
One honest note: there is no universal multiplier. The outcome depends on your industry, your ticket, and how much of your volume is urgent versus price-shopping. Anyone promising a fixed return is selling, not calculating.