How to Export Sora Videos Before the API Shutdown

Photo by Peter Stumpf on Unsplash

You can export Sora videos before OpenAI shuts down the remaining API path on September 24, 2026, either through OpenAI’s Sora sunset export page or by downloading completed API videos as MP4 files. This is useful if you created videos in the discontinued Sora web or app experience, or if your team still has generated assets sitting behind the Videos API.

The deadline is not cosmetic. OpenAI’s Sora discontinuation page says the Sora web and app experiences were discontinued on April 26, 2026, and the Sora API will be discontinued on September 24, 2026.

The shortest safe path to export Sora videos

To export Sora videos from the consumer Sora experience, use OpenAI’s sunset export page, request the export, wait for the ready email, and download the archive before the export window closes. For API-created videos, list your generated videos, identify completed jobs, then download each final MP4 through the Videos API content endpoint before September 24, 2026.

1. Open sora.chatgpt.com/sunset while signed in to the account that created the Sora content.
2. Click Export on the sunset page.
3. Watch for the email OpenAI sends when the export is ready.
4. Download the export immediately rather than treating the email as long-term storage.
5. For API assets, call GET /videos, then download completed videos with GET /videos/{video_id}/content.
6. Store the MP4 files outside OpenAI before deleting anything from OpenAI storage.

OpenAI’s Sora help article gives the consumer export path directly: go to sora.chatgpt.com/sunset, click Export, and wait for the readiness email. OpenAI’s video generation guide gives the API side of the same rescue job: completed videos can be downloaded as final MP4 files through GET /videos/{video_id}/content, while GET /videos lists videos with optional pagination and sorting.

Decide which Sora content you are actually trying to save

Two shutdowns matter here, and they affect different people. OpenAI says the Sora web and app experiences ended on April 26, 2026, while the Sora API remains on a separate deadline of September 24, 2026.

That split creates the easiest place to make a bad assumption. If you made clips through the old web or app experience, the relevant path is the sunset export page. If your product, internal tool, or automation generated clips through the API, the relevant path is the Videos API, which OpenAI’s developer guide says is deprecated and scheduled to shut down on September 24, 2026.

Teams should treat those as two inventories, not one archive. The consumer export flow is account-centered, while the API flow is object-centered: you need video IDs, completion states, and a download pass for each final asset. That is the non-obvious trap in this shutdown, because an email export can make a creator feel finished while API-generated MP4s remain untouched in a different workflow.

If you landed here after searching for Sora discontinued export data, the only official consumer instruction OpenAI gives is still the sunset export path. OpenAI’s help article also says that after Sora is discontinued and any final export window passes, data associated with a user’s Sora use will be permanently deleted.

Save web and app creations through the sunset export page

Start with the browser flow if your Sora work came from the discontinued web or app experience. OpenAI says users can export content created in Sora by going to sora.chatgpt.com/sunset and clicking Export, then waiting for an email when the export is ready.

Use the same OpenAI account that created the Sora content. OpenAI’s published instruction does not describe a cross-account migration path, so the safest interpretation is plain: the export request should come from the account that owns the Sora history. Do not ask a teammate to request an export unless that teammate’s account is the one that created the assets.

Once the email arrives, download the export and move it to storage you control. OpenAI’s shutdown note says deletion comes after Sora is discontinued and after any final export window passes, if one is offered, so the export email should be treated as a pickup notice rather than an archive guarantee. Waiting is the mistake.

A clean export folder should separate raw downloads from edited campaign files, especially if Sora clips were used in client work, ads, or social posts. The article belongs beside AnIntent’s AI Video articles because the task is not just file recovery; it is continuity for creative work that may have moved across several tools after generation.

Download API-created Sora videos before the endpoint disappears

The API path is more mechanical. OpenAI’s video generation guide says users can list videos with GET /videos, using optional pagination and sorting parameters, then download completed videos as final MP4 files with GET /videos/{video_id}/content.

That is the answer to how to download Sora videos if your clips were generated through code rather than the old Sora interface. List the videos first, filter for jobs that completed, then download the MP4 content for every video ID you need to keep. A spreadsheet with video_id, status, project owner, and local filename is dull, but it prevents duplicate downloads and missed assets when multiple developers used the same project.

OpenAI also says video generation is asynchronous: POST /videos returns a job object with an ID and status, and users can poll GET /videos/{video_id} or use webhooks until the job completes, according to OpenAI’s video generation guide. That matters for export work because a job ID is not the same thing as a finished MP4. The file exists only after the job reaches a completed state.

For a small archive, a manual loop is enough: call GET /videos, download each completed asset, and record the result. For a larger archive, script the process so pagination does not silently hide older jobs. OpenAI’s guide confirms pagination exists for listing videos, but it does not promise that one list response will contain every historical item.

The Sora API shutdown September 24 2026 also affects model names, not just one generic endpoint. OpenAI’s video generation guide lists the affected API and model names as Videos API, sora-2, sora-2-pro, sora-2-2025-10-06, sora-2-2025-12-08, and sora-2-pro-2025-10-06.

The failed download that usually means the job is not complete

The most common export failure is trying to fetch video content from a job that has not finished. OpenAI’s guide describes Sora video generation as asynchronous, with POST /videos returning a job object and users polling GET /videos/{video_id} or using webhooks until completion.

The fix is not to retry the MP4 download harder. Check the job status first, then download only after the job is completed. If your script treats every video_id as a finished file, it will mix real download errors with ordinary waiting states, which is a bad failure mode on the last day of a shutdown.

A practical export script should create three buckets: completed and downloaded, pending or still processing, and failed or no longer needed. OpenAI’s video generation guide says users can delete videos they no longer need from OpenAI storage with DELETE /videos/{video_id}, but deletion should come after you have verified the MP4 is stored elsewhere. Delete too early and the cleanup task becomes the data-loss event.

Check model-specific leftovers before closing your archive

Sora 2 and Sora 2 Pro are both legacy paths in OpenAI’s current documentation. OpenAI’s Sora 2 model page describes Sora 2 as a legacy default model for flagship video generation with synced audio, and it lists support for text input only, image input only, audio output only, and video output only.

The Pro model deserves a separate sweep because teams often used higher-cost models for client-facing output. OpenAI’s Sora 2 Pro model page describes Sora 2 Pro as a legacy default model and OpenAI’s most advanced synced-audio video generation model. If your export inventory tags model names, sort by sora-2-pro before you call the job done.

Pricing gives another clue about which assets deserve review first. OpenAI’s Sora 2 page lists Sora 2 video generation at $0.10 per second for 720x1280 portrait or 1280x720 landscape output. OpenAI’s Sora 2 Pro page lists Sora 2 Pro at $0.30 per second for 720x1280 or 1280x720 output, $0.50 per second for 1024x1792 or 1792x1024 output, and $0.70 per second for 1080x1920 or 1920x1080 output.

Those historical prices do not tell you what a clip is worth now, but they help rank the archive. A 12-second Sora 2 Pro job at 1080x1920 represented a materially different generation choice than a 4-second Sora 2 job at 720x1280, according to OpenAI’s listed per-second model pricing. When time is short, export the highest-resolution Pro outputs first, then work down to experiments and duplicates.

Stop creating new Sora dependencies while exporting old ones

OpenAI’s API reference still documents video creation, but it marks Create video as deprecated and says the Sora API is scheduled to permanently shut down on September 24, 2026, according to OpenAI’s TypeScript API reference. That same reference lists sora-2 and sora-2-pro as allowed video generation models for the create endpoint, with sora-2 as the default.

If you are still generating final work on September 23, stop treating Sora as a production dependency and treat it as an export source. OpenAI’s create video reference lists allowed clip durations of 4 seconds, 8 seconds, or 12 seconds, with 4 seconds as the default, and it says the prompt field is required with a minimum of 1 character and a maximum of 32,000 characters. Those details explain why old prompts and job metadata can still matter during archive review, even if the endpoint is no longer a sensible place to start new work.

Reference inputs deserve special attention. OpenAI’s create video reference lists input_reference as an optional reference asset upload or reference object that guides generation. If your audit only downloads final MP4s, you may preserve the finished video while losing the context that explains why a particular clip looks the way it does.

This is where export discipline crosses into reproducibility. Save final MP4s first because OpenAI explicitly documents those as the completed video download format, then preserve prompts, video IDs, model names, durations, and any internal campaign notes your system stored. AnIntent’s Developer Tools articles often return to the same lesson: generated media pipelines fail messily when metadata is treated as less valuable than output files.

Credits are not the same thing as saved files

Do not confuse account credit questions with export completion. OpenAI’s Sora discontinuation page says purchased ChatGPT/Sora credits can still be used for Codex, so the shutdown does not necessarily mean every paid credit becomes unusable.

That does nothing for an undownloaded Sora MP4. Credits can shift to another OpenAI product path, but the Sora export problem is about files, metadata, and deletion timing. If you have to choose between checking credit balances and downloading completed videos, download first.

For organizations, the practical owner should be whoever controls the API project and storage destination, not whoever paid the invoice. Product, marketing, and legal teams can argue about retention policy after the MP4s are safely outside the deprecated service. The clock does not care which department owns the budget.

Archive the result so the shutdown does not create a second loss later

A rushed export can still fail six months from now if filenames make no sense. Use filenames that preserve at least the project name, date, model, orientation, and video ID where available. OpenAI’s API docs identify video_id as the key used to poll and download video content, so carrying that ID into your local filename gives you a durable bridge between logs and files.

Store the exported MP4 files in two places if the videos matter to active work. One copy should be in the creative team’s normal asset library, and another should be in colder storage with limited write access. That is not an OpenAI-specific rule; it is basic media hygiene made urgent by a dated shutdown.

After the MP4s are verified, decide whether to remove unneeded OpenAI-stored copies. OpenAI’s video generation guide says DELETE /videos/{video_id} can delete videos from OpenAI storage, while OpenAI’s Sora discontinuation page says Sora-associated data will be permanently deleted after discontinuation and any final export window passes. If your organization has retention rules, export first, verify second, delete third.

Creative teams tracking AI media workflows should also watch AnIntent’s AI Tools articles and AI Industry articles for shutdowns that turn product strategy into user data chores. The immediate next step is narrower: request the sunset export now, then run the API download pass for every completed video ID before September 24, 2026.