OpenAI announced that the Sora web and app experiences shut down on April 26, 2026, and the Sora API follows on September 24. Sora 2 was, for a while, the best physics model in the business — the one that got water, fire and gravity right when everything else looked like a screensaver.

And it’s being switched off.

If you built your content pipeline on the Sora app — your prompts, your saved generations, your team’s muscle memory — you now have a migration project. Pick a new model. Learn a new UI. Rebuild your prompt library. Re-key your billing. Hope the thing you replace it with is still around next year.

This keeps happening

Sora isn’t special. Models get deprecated, rebranded, price-hiked, rate-limited, and occasionally just turned off. The half-life of “the best model” in 2026 is measured in weeks. Betting your workflow on any single provider is betting that the fastest-moving industry in the world will sit still for you. It won’t.

The mistake isn’t picking Sora. Sora was a great pick in its moment. The mistake is building anything durable on top of exactly one vendor’s roadmap.

What “not locked in” actually looks like

In Any AI Studio, video generation isn’t “the Sora feature.” It’s a composer that routes to whichever model fits the shot:

  • Veo 3.1 when you need synced audio and cinema-grade color.
  • Kling 3.0 for the best motion-to-cost balance at volume.
  • Seedance 2.0 when you’re generating a lot and want the quality-per-dollar leader.
  • Runway Gen-4.5 for tight cinematic control, Wan 2.6 when you need a clip in twenty seconds.

When Sora’s API goes dark in September, users who generated through it in the studio will do exactly one thing: pick a different model from the same dropdown. No migration. No re-keyed billing. No lost history. The prompt they wrote still works, because it was never coupled to one vendor’s endpoint in the first place.

The subscription you actually want

The pitch for a single-provider subscription is simplicity: one app, one bill. The problem is that it’s simple right up until the provider changes the deal — and in this market, the provider always changes the deal.

An aggregator gives you the same one-app, one-bill simplicity, minus the single point of failure. New model launches? It shows up in the picker. Old model sunsets? It quietly drops out, and your work carries on with its replacement.

Sora going away is sad. It was genuinely good. But it’s also the clearest argument we’ve seen all year for not marrying any one model — and for keeping every frontier option one click apart instead.