Google sets 30 September 2026 deadline to drop gemini-omni-flash-preview

If your application calls gemini-omni-flash-preview, you have a firm deadline to work to. Google has published a hard deprecation date of 30 September 2026 for that endpoint. After that date, calls to the preview model will fail. The replacement is gemini-omni-1.1-flash, which reached general availability on 27 August 2026.

With roughly four weeks between the GA launch and the cutoff, the window for testing and migrating is short but workable if you start now.

What changed and why it matters

gemini-omni-flash-preview launched on 30 June 2026 as a public preview of Google’s conversational video generation model. Preview endpoints are explicitly not guaranteed stable: Google’s own policy notes they ship with potentially tighter rate limits and can be deprecated with as little as two weeks’ notice.

The GA release, gemini-omni-1.1-flash, is the production-grade version of the same model. Alongside the stable identifier, it brings new capabilities that were not in the preview:

• Video extension. The model can now continue an existing clip, analysing up to 10 seconds of prior footage and appending extensions in 10-second increments up to 40 seconds cumulative. Extensions are chained via a previous_interaction_id parameter, so this is a stateful editing session rather than a fresh re-prompt.
• Interpolation between two images. Using the image_to_video task with up to two images, the model generates a video that transitions between the first and last frames.
• Resolution control. A new resolution parameter in video_config accepts 360p, 720p (the default), 1080p, and 4K. The two higher options are produced via upscaling rather than native rendering.

The model runs on the Interactions API rather than generateContent, which is what enables multi-turn video editing: generate a clip, then send a follow-up turn that edits the result without re-uploading source material.

What you need to do before 30 September 2026

The core migration step is straightforward: swap gemini-omni-flash-preview for gemini-omni-1.1-flash in your model string. That one-line change is usually enough to keep existing text-to-video and image-to-video flows running.

However, there are a few things worth validating before you treat the migration as done.

Check your resolution handling. The default output is now 720p, and 360p and 4K are newly available options. If your code was written against the preview and assumes a fixed output size or resolution, confirm what you are actually receiving. The specification may have drifted from what your integration expects.

Handle large video responses correctly. Videos over 4 MB are returned as a URI rather than inline base64. If your response handler only reads output_video.data, a higher-resolution output will silently return nothing. Confirm your handler checks for both response shapes.

Choose the right task ID. The GA model page documents distinct task-specific identifiers: text-to-video, image-to-video, reference-to-video, video-edit, and video-extend. Make sure you are using the correct one for each flow in your application.

Validate quality, cost, and latency with your own assets. Switching model strings is fast; validating that your specific prompts, input images, and output-review process behave as expected takes longer. Budget time for that, especially if you are generating at 1080p or 4K where upscaling behaviour may differ from what you saw at 720p in preview.

Review your monitoring and error handling. Entitlements, rate limits, and error response shapes can differ between preview and GA tiers. Run your standard monitoring checks after the switch, not just a functional smoke test.

A note on cost

There is no free tier for Gemini Omni. Every generation is billed, and that was true of the preview as well. One practical option during testing is to use 360p output, which generates up to 60% faster and costs roughly a third of 720p output. That makes it a useful draft mode for validating prompts and flows before committing to full-resolution generation.

Where Omni 1.1 Flash fits in the broader model picture

It is worth being clear about what this model is and is not. gemini-omni-1.1-flash is Google’s conversational video model: you generate a clip, then refine it across turns. Veo 3.1 is a separate model aimed at cinematic rendering with native audio. Nothing in this deprecation or GA release affects Veo integrations.

Google has also noted named integration partners for Omni including Adobe Firefly, Figma Weave, Runway, and GMI Cloud, which suggests this endpoint will sit beneath a growing number of creative tools. If you are building on top of any of those surfaces rather than calling the API directly, check with the relevant platform about their own migration timelines.

The short version

Audit your codebase for references to gemini-omni-flash-preview, update to gemini-omni-1.1-flash, test against your own prompts and assets at your target resolution, and verify your response handler covers both the inline and URI response shapes. The Gemini API changelog and Omni documentation have the full technical details. The deadline is 30 September 2026, and there is no grace period beyond it.