Why Sora ended, facts only

OpenAI discontinued the Sora web and app experiences on 26 April 2026. It will discontinue the Sora API on 24 September 2026. Those two dates are from OpenAI's own help article. The help article does not state a reason. Everything below that is a motive is attributed to a named outlet with a date. If a source is not cited, it is not a fact on this page.

This is not a eulogy and it is not a routing table. The Sora API sunset migration plan is the checklist. GPT-6 Astra will not make your videos is the adjacent mix-up: Astra is not a Sora heir. This page is only what can be sourced about the end of the product.

What OpenAI published

The help centre article, titled "What to know about the Sora discontinuation," states:

• The Sora web and app experiences were discontinued on 26 April 2026.
• The Sora API will be discontinued on 24 September 2026.
• Users can export content at sora.chatgpt.com/sunset. OpenAI recommends exporting as soon as possible. If a final export window is offered, OpenAI says it will notify by email before it begins.
• After discontinuation, and after any final export window, OpenAI says it will permanently delete data associated with use of Sora.
• Purchased ChatGPT/Sora credits can still be used for Codex.

That is the official object. Two dates, an export path, a deletion statement, a credits note. No paragraph on cost, users, Disney, or strategy.

OpenAI announced the discontinuation on 24 March 2026. Contemporaneous coverage of that announcement — including PCMag on 30 March 2026 — records that OpenAI's public line was thin: a goodbye to the video-generation app, and a later statement that the company faced growing compute needs and would shift the Sora research team to "world simulation research to advance robotics that will help people solve real-world, physical tasks." That quoted phrase is OpenAI's, as reported. It is a statement about where the team would go. It is not, by itself, a cost model.

Wikipedia's Sora article, citing the same period, notes that OpenAI did not provide a specific reason in the shutdown notice. That matches the help page you can read today.

The catalog still lists Sora 2 Text to Video and Sora 2 Pro Storyboard until the API is gone. Treat those cards as a calendar, not as a promise.

What contemporaneous reporting attributed

On 29–30 March 2026, several outlets reported on a Wall Street Journal investigation into the shutdown. This page repeats what those outlets attributed to the Journal. It does not independently verify the Journal's numbers.

TechCrunch, 29 March 2026, citing the Journal:

• After launch, Sora's worldwide user count peaked at around a million and then collapsed to fewer than 500,000.
• The app was "burning through roughly $1 million every day" because video generation is costly to run.
• TechCrunch's summary of the Journal piece is that Sora "was a money pit that nobody was using, and keeping it alive was costing OpenAI the AI race."
• TechCrunch, still citing the Journal, reported that Disney — which had committed $1 billion to a partnership — "found out Sora was being shut down less than an hour before the public."

PCMag, 30 March 2026, also citing the Journal:

• "Sora had quietly turned into a liability for OpenAI in the months after its release, particularly as the startup tightened its focus ahead of a looming IPO."
• App download figures in that piece: 4.8 million in October, 6.1 million in November, then 1.1 million "last month" (February 2026 in the reporting), "but still cost OpenAI $1 million a day."
• PCMag records OpenAI's compute-and-robotics statement, quoted above.

The Decoder, 30 March 2026, again pointing at the Journal, repeats the ~$1 million a day figure, the peak-around-a-million then ~500,000 user path, and the redirect of compute toward coding, enterprise, and agent products. The Decoder also reports, as part of that coverage, that OpenAI canceled training runs for new video models. That claim is the Decoder's report of the Journal story. It is not an OpenAI help-centre fact.

Where a later recap names $2.1 million in lifetime revenue, treat it as a figure that has circulated in secondary write-ups of the same episode. This page does not treat that number as something OpenAI published. If you need it in a deck, cite the outlet you actually opened, not this sentence.

None of those outlets are OpenAI. The $1 million-a-day compute cost is a Journal figure as relayed in March 2026 coverage. Attribute it that way. Do not restate it as "OpenAI said the app cost $1 million a day." OpenAI, in the materials above, did not.

What OpenAI did not say

The help article does not say:

• that Sora failed
• that video is unprofitable in general
• that a successor model is coming
• that ChatGPT will keep a production-grade Sora API after 24 September
• that export will remain available indefinitely

It also does not say that GPT-6 Astra, announced 3 September 2026, generates video. Astra is a different product. Do not migrate pixels to it.

Absence is a fact too. A production plan that assumes a hidden Sora endpoint, a grace generate after the 24th, or a like-for-like OpenAI video API is a plan that is not in the help centre. OpenAI's own words on data are the opposite of a grace window: export, then deletion after any final window.

What you can still do until 24 September 2026 is call the API, if your integration still works. That is not advice to start new work there. It is the remaining clock the help page named.

What a production team can actually do with this

You do not need a theory of OpenAI's board to finish the week.

Believe the two dates. App already gone. API on 24 September 2026. Put the second date on a calendar that is not a Slack rumour.

Export. Use the sunset URL OpenAI published. Do it before you argue about replacements. A model change will not resurrect a file you did not download. OpenAI says it will delete Sora-associated data after the process it described.

Do not convert Journal cost figures into your budget. $1 million a day is a reported OpenAI-side infrastructure number. It is not a price you were paying, and it is not a Versely credit. Your cost to replace the work is the catalog row you actually run.

Route the jobs, not the brand. Sora was a name over dialogue, longer identity-locked takes, motion, and talking shots. Those already live as separate rows. The AI video generator is the launcher. Veo, Seedance, and Kling are live generators. Astra is not on that list.

Leave residual ChatGPT generation out of production. If a Sora-like button still appears inside a chat product after the API dies, it is not an API, not a batch path, and not a date you control. The help page you can cite is about the app already being gone and the API shutting on 24 September.

Open a studio account at Sign up when the next action is generating on a live row, not waiting for OpenAI to publish a reason it already declined to put on the help page.

The facts that matter for shipping are short: two dates, one export URL, one deletion statement, and a set of reported cost and usage numbers that belong to the Journal's March 2026 story. Everything else is commentary. This page stops before commentary.

FAQ

Did OpenAI say Sora ended because it cost $1 million a day?

No. That figure comes from Wall Street Journal reporting as relayed by TechCrunch and PCMag in late March 2026. OpenAI's help article lists dates, export, deletion, and credits. It does not list a daily compute cost.

Is the API already dead?

Not as of this post's date. The help centre still names 24 September 2026. The consumer web and app experiences already ended on 26 April 2026. Do not collapse the two dates. The one that breaks scripts is the second.

Will Astra replace Sora?

OpenAI has not announced video generation for GPT-6 Astra. Treat "Astra will make the clips" as a mix-up, not as a migration. Pixel jobs go to live video models. The Astra post on this site is explicit about that boundary.