Should you wait for Veo 4? Build on a swappable model id instead

Do not wait. As reported in a PixVerse article summary read on 2026-10-02, Veo 3.1 was still Google's production Veo as of 2026-09-20, with no official Veo 4 date, model id or price page. Without a date, waiting has no end, so ship on a model you can call now, keep the model id in config, and treat a Veo 4 launch as a swap you test, not a deadline.

That Veo status is reported by PixVerse, whose page I saw only as a search-result summary, so I treat it as a report and not as a Google statement. Sume's side comes from the video generation docs: the video catalog on main lists no Veo id, so a Veo model would be a new row, not a rename.

What is known and not known about Veo 4?

Keep the two lists apart so nobody schedules a launch around a rumor.

What can you build on today?

Sume's video catalog already lists models across several vendors. Gemini Omni Flash 1.1 (gemini-omni-flash-1.1) takes text, image and reference-video inputs, makes 3 to 10 second clips from 360p up to 4K in 16:9 or 9:16, and always includes audio. Other rows such as kling-3 and seedance-2.5 cover other lengths and ratios. The right pick is the row whose limits match your brief.

What makes a model swap cheap?

Four habits:

• Read the model id from config or an environment variable, never from a string in the request builder.
• Validate duration, resolution and aspect_ratio against the catalog response, not constants.
• Store the prompt, the model id and the output with each job, so a later A/B is possible.
• Keep a five-prompt regression set and re-run it when a new id appears.

When would waiting be reasonable?

If your project needs something no listed model can do, such as a length or a ratio outside every catalog row, then waiting for a new model is a real option. Otherwise the cost of waiting is every clip you did not make, and a swap later costs one config change and one regression run.