OpenAI retires the Sora API: the September 24 cutoff arrived with no replacement model listed

Start with the table from the official page, because it is clearer than any retelling. The entries and dates listed under this deprecation are: Videos API shut down on 2026-09-24 with an empty recommended replacement; sora-2 on 2026-09-24, empty; sora-2-pro on 2026-09-24, empty; sora-2-2025-10-06 on 2026-09-24, empty; sora-2-2025-12-08 on 2026-09-24, empty; and sora-2-pro-2025-10-06 on 2026-09-24, empty. In other words, there is no line in the documentation that tells you which model to send your requests to instead.

This matters because other deprecations on the same page generally do name a successor. Another section explains that a set of older, declining-usage models will be shut down over the next six to twelve months, with dates including September 28, 2026 for gpt-3.5-turbo-instruct and babbage-002, both pointing to gpt-5.6-terra as the recommended replacement. Further up, in the fine-tuned model list, ft-gpt-3.5-turbo, ft-gpt-4, ft-gpt-4.1-nano-2025-04-14, ft-babbage-002 and ft-davinci-002 shut down on October 23, 2026 and map to replacements such as gpt-5.6-terra, gpt-5.6-sol and gpt-5.6-luna. Same maintainer, same table: some rows offer a path forward, the Sora rows do not.

The second thread is timing. The official page gives two points: notification on March 24, 2026 and removal on September 24, 2026, exactly six months apart. As for the consumer product, a post in the OpenAI Developer Community asking the company to release Sora 2 weights before shutdown cites OpenAI's discontinuation notice as saying the Sora web and app experiences were discontinued on April 26, 2026, while the API is scheduled to be discontinued on September 24, 2026. Note the difference in sourcing: the notification date and the API shutdown date come from OpenAI's official deprecations page, whereas the April 26 line comes from a developer community post citing the official notice. This article does not merge the two.

The third thread is how the deprecation propagates downstream. The Make help center states that because of OpenAI's decision to deprecate its entire Videos API and all Sora models, Make is deprecating all video-related modules in its OpenAI app, with an action required by September 24, 2026: after that date, scenarios using the deprecated video modules will stop running, and new scenarios using those modules can no longer be created. The same page states that OpenAI has not announced a replacement. This is what an upstream deprecation looks like downstream: a node in your automation that calls video generation does not merely get slower in some version, it simply stops working.

The fourth thread is migration, and here the facts must be kept apart from advice. The fact is that, per OpenAI's official deprecations page, no replacement model is offered for these rows. At the advice level, third-party write-ups are looking to other vendors' video models; one comparison article places Sora 2 alongside Google's Veo line and xAI's Grok Imagine and notes that pipelines built on the Sora API need to be rebuilt after September 24, 2026. That is third-party analysis, not an OpenAI statement, and the two should not be cited interchangeably. For teams that genuinely need to migrate, the verifiable next steps are only three: inventory which calls hit the Videos API, pin down the hard requirements those calls have on video quality, audio and duration, and run the same prompt set through a few candidate models to compare.

One closing note on the size of this story. It does not come with a launch event or a demo reel, but its impact is certain: any team that put video generation into a production workflow has to change code starting today. And the detail that the official table offers no replacement is worth remembering more than the shutdown itself, because it means the migration path is yours to define rather than the documentation's.