Omni Flash 1.1: GA in the Gemini API, Preview on Agent Platform

Two Google pages use different wording. The Gemini API release notes record general availability for gemini-omni-1.1-flash on August 27, 2026. The Google Cloud Agent Platform page for the same model is titled "Gemini Omni 1.1 Flash Preview". The body text fetched for this post was mostly navigation, so it gave no regions or quotas. Treat the route you call, not the model name, as the thing to verify.

What each page says

The models page lists Omni Flash under the id gemini-omni-1.1-flash with no preview suffix, while the Veo 3.1 row uses veo-3.1-generate-preview and is labelled Preview. The release notes say on August 27 that Omni went GA as gemini-omni-1.1-flash, adding extension, interpolation and resolutions from 360p to 4K, and that the earlier preview endpoint (gemini-omni-flash-preview) is deprecated on September 30, 2026.

The Agent Platform page carries the word Preview in its title. The fetched text was mostly navigation, so no limits were extracted from it.

Why it matters in a contract

A Preview label on a cloud platform can mean different support terms from a GA API model, but the page text read here did not say which terms apply, so do not assume. If a procurement checklist asks for GA-only models, ask Google for the written status of the exact route you plan to use, and save the page date.

Also note the Gemini API route runs through the Interactions API, not generateContent, per the Omni guide.

What Sume exposes

The Sume docs read for this post do not label models Preview or GA. Its Video Router offers gemini-omni-flash-1.1 (note the word order differs from Google's id) with 3 to 10 second output at 360p, 720p, 1080p or 4K, 16:9 or 9:16, and native audio. A single request shape covers text, image, reference and edit modes, and billing is provider list times 1.25 per output second.

If your question is only whether the API behaves as documented, a quick test on a small clip answers it faster than reading status labels. Run one 3 second 360p generation and check the returned job and artifact before you plan a campaign.