OPENAI · SORA SHUTDOWN
Sora shuts down its API too, in September.
Under the two-stage shutdown plan OpenAI announced in March 2026, the Sora app and web experience already closed on April 26. The developer-facing Sora 2 API will stop working on September 24, 2026.
Six months, and both the app and the API are gone
Why did the product that was OpenAI's face for video generation get wound down this fast?
The Sora app, released in late September 2025 to viral buzz, peaked at roughly 3.3 million monthly downloads in November 2025 before collapsing to about 1.1 million by February 2026. On March 24, 2026, OpenAI formally announced Sora's phased discontinuation on its official Help Center. The web and app experiences shut down first, on April 26, 2026, and the developer-facing Sora 2 API (including sora-2, sora-2-pro, and the Videos API) is set to shut down completely on September 24, 2026.
As of today, August 22, 2026, that leaves roughly 33 days before the API goes dark. If your app or automation pipeline assumes Sora, now is the time to pick a replacement and start testing it. Forbes frames the episode as a textbook case of how fragile a workflow becomes when it depends on a single AI vendor.
The bottom line: the economics never worked
The gap between operating cost and revenue is reported alongside a sharp collapse in users.
OpenAI hasn't disclosed exact figures, but multiple outlets report operating costs of roughly $1 million a day (with some estimates reaching $15 million a day at peak) against total lifetime app revenue of only about $2.1 million (the-decoder.com). Adding to the pressure were backlash over copyright and deepfake concerns, plus the collapse of a planned three-year, $1 billion character-licensing deal with Disney — which reportedly learned of the shutdown less than an hour before it was announced.
What to do before September 24, by role
The impact goes beyond simply losing access to video generation.
Designers & creatives
If you used Sora for short-form social clips or visual prototyping, generation stops on September 24. Download your finished assets from the Sora library now, and start testing whether Veo 3.1 or Runway Gen-4.5 can reproduce your look and feel.
Marketers
Teams that built Sora into creative-production pipelines lose that supply on September 24. Audit your existing video assets now and move up your test of a lower-cost alternative like Kling 3.0, ahead of your campaign calendar.
Engineers
Calls to sora-2 / sora-2-pro will start erroring out on September 24. Swap your Videos API endpoint for Veo 3.1 or the Runway API, and verify request-shape differences such as webhook support (Sora is poll-based; Runway and Kling support webhooks) before you cut over.
Product leads
If video generation was a core feature, this is worth sharing internally as a case study in single-vendor risk. For your next vendor selection, build service continuity and data-migration terms into the contract — not just price.
Replacement options and rough pricing
There's no single drop-in replacement — the right pick depends on your use case.
| Sora 2 API (shutting down) | Rough migration targets |
|---|---|
| $0.10–0.70/sec (varies by report) | Veo 3.1: from $0.15/sec in fast mode |
| No webhooks — polling only | Runway and Kling both support webhooks |
| All models end Sep 24, 2026 | Kling 3.0 sits in the ~$0.10/sec tier |
| No successor model announced | Runway Gen-4.5 leans toward editing-led workflows |
Prioritize fidelity and you'll likely land on Veo 3.1; prioritize cost and Kling 3.0 wins; prioritize editing-centric workflows and Runway Gen-4.5 fits best — a split that developer comparisons converge on as well.
A practical checklist before September 24
Back up your assets
Download every video and image from the Sora library into your own storage — the export feature only works while the service is still live.
Pick a replacement and run a PoC
Trial Veo 3.1, Kling 3.0, or Runway Gen-4.5 at small scale for your use case, and compare quality, latency, and price.
Cut over your pipeline
Rewrite your integration for differences like webhook support and response shape, and complete the production cutover before September 24.
What teams depended on wasn't the AI itself —
it was one vendor's staying power.
Reasons not to read too much into this
The cost and revenue figures here aren't official OpenAI disclosures — they're media estimates, and they vary considerably (some reports cite roughly $1 million a day, others put the peak as high as $15 million). What matters more than the exact number is the underlying structural point: video generation costs far more to run than text or image generation, an industry-wide problem, not one unique to OpenAI.
It's also premature to conclude OpenAI is exiting video generation entirely. Forbes's analysis frames the decision as part of a broader strategic shift — reallocating compute toward coding and enterprise products ahead of an expected IPO. A return to video generation, under a different form, can't be ruled out.