Sora API Shutdown Sept 24: The Final Developer Checklist
Exact affected endpoints, migration targets, and the option in OpenAI's own docs that most guides never mention.
Hard Deadline
September 24, 2026. OpenAI removes the Videos API, sora-2 and sora-2-pro. Developers were notified March 24, 2026. There is no OpenAI replacement model.
If anything in your stack calls POST /v1/videos, you have a hard cutover coming and no vendor-supplied landing pad.
That second part is what makes this different from a routine model deprecation. OpenAI's deprecations page has a recommended-replacement column, and for most retired models it names a successor. For the Videos API and Sora 2 row, that column is empty. Every migration path from here leads to a different vendor, with a different API shape, different prompt behaviour and different output characteristics.
Here is the complete checklist, in the order you should work through it.
📋 The Checklist
Two Dates People Keep Confusing
This is the single most muddled point in circulation, and getting it backwards sends you chasing a deadline that expired months ago.
| Date | What Ended | Who It Affected |
|---|---|---|
| April 26, 2026 already passed |
Sora web and app experiences, including sora.chatgpt.com | Consumer users |
| September 24, 2026 upcoming |
The Videos API and all Sora 2 model IDs | Developers |
The API has continued operating normally since the app shutdown. If your integration is still running today, that is expected — not a sign the deadline moved.
Exactly What Is Being Removed
OpenAI's deprecations page names the affected surfaces precisely. On March 24, 2026, developers using the Videos API and Sora 2 model aliases and snapshots were notified of their deprecation and removal from the API on September 24, 2026.
In practical terms, grep for all of these:
# Endpoint
POST /v1/videos
GET /v1/videos/{video_id}
/v1/videos # any variant, including polling and download paths
# Model IDs
sora-2
sora-2-pro
# plus any dated snapshot aliases you pinned
# SDK symbols (deprecation annotations shipped Aug 13, 2026)
client.videos.*
Videos, VideoService, VideoCreateParams # naming varies by SDKOpenAI's official SDKs began carrying deprecation annotations in August 2026, applied across blocking, async, raw-response, overload, parameter and pagination surfaces. If you upgraded your SDK recently, your build logs are probably already warning you.
Step 1 — Audit Your Codebase
Do this first and do it broadly. Video generation calls hide in places you will not expect — scheduled jobs, queue workers, admin tools, an internal Slack bot somebody built and forgot.
# Cast a wide net across your repos
grep -rn --include="*.{py,js,ts,jsx,tsx,go,rb,java,php}" \
-e "sora-2" \
-e "v1/videos" \
-e "\.videos\." \
-e "VideoCreate" \
.
# Don't forget these
# - environment configs and feature flags
# - CI/CD pipeline definitions
# - no-code scenarios (Make, n8n, Zapier)
# - database rows storing model IDs as stringsThat last one catches people. If you persisted sora-2 as a string in a jobs table or a user preferences record, a code-only grep will miss it and the failures will surface at runtime after the cutoff.
Step 2 — Export Before Deletion
OpenAI has stated that after Sora is discontinued — and after any final export window passes, if one is offered — all data associated with your use of Sora will be permanently deleted.
The export path: go to sora.chatgpt.com/sunset, click Export, and you will receive an email when the archive is ready.
⚠️ Do not wait for a final export window. OpenAI's language is conditional throughout — it says it will notify users by email if a final window is offered, and that it had not decided whether one would happen. Treat the export as due now, not as a task with a guaranteed grace period behind it.
Separately, if your application stores generated video URLs rather than the files themselves, download the actual assets. Signed URLs from any provider expire, and yours will not outlive the service.
⚡ Before You Start Rewriting Against Another Vendor
There is one line on OpenAI's deprecations page that almost no migration guide mentions, and for some teams it changes the entire plan.
Read the next section before you commit engineering time.
Step 3 — The Option Most Guides Miss
Buried in the general text of OpenAI's deprecations page, above the model tables, is this: in some cases developers may be able to provision dedicated capacity for continued access after a model's shutdown date, and teams interested in that option are directed to contact OpenAI's sales team.
This is not advertised in the Sora-specific announcements, and virtually every migration article omits it entirely.
Two honest caveats. It says "in some cases," which means it is discretionary rather than guaranteed. And dedicated capacity is an enterprise commercial arrangement — realistically relevant if you have meaningful volume, not if you generate a few clips a week.
But if you run Sora in production with a hard dependency and twenty days is not enough time to migrate safely, one email to OpenAI sales costs you nothing and is a considerably better use of an afternoon than a rushed cutover. Ask the question before you assume the answer.
Step 4 — Pick Your Migration Target
Since OpenAI names no successor, choose by what you actually used Sora for rather than by general quality rankings.
| If You Used Sora For | Closest Target | Why |
|---|---|---|
| General text-to-video with audio | Google Veo 3.1 | Widely cited as the closest feature parity — native audio, 1080p, vertical 9:16, strong prompt fidelity |
| High volume on a tight budget | Kling 3.0 | Generally the cheapest credible replacement per generation |
| Editing-led production workflows | Runway Gen-4.5 | Built around iterative editing rather than one-shot generation |
| Short stylised social clips | Seedance 1.0 Pro | Named by aggregator platforms as a direct sora-2 substitute in their catalogues |
Budget for prompt rewriting, not just endpoint swapping
The API call is the easy part. Prompt syntax does not transfer cleanly between video models — camera direction, scene composition and motion vocabulary are interpreted differently by each. A prompt library tuned over months against Sora 2 will not produce equivalent output elsewhere without rework.
Plan for the full surface: prompt libraries, asset storage paths, moderation behaviour, retry and rate-limit handling, and output validation. Teams consistently underestimate this because the endpoint change looks trivial.
Why Third-Party Proxies Are Not a Safety Net
Several aggregator platforms integrated the Sora API and continued offering access after the consumer app closed. It is tempting to treat one of these as a buffer.
Do not. At least one major aggregator sunset its sora-2 route in late April 2026 — five months ahead of OpenAI's own date — and returns HTTP 410 with a structured error body for requests to it now.
A reseller's access ends whenever that reseller decides it ends, which can be earlier than the upstream date and typically comes with less notice. Routing through a proxy converts one known deadline into an unknown one.
The Day-by-Day Plan
Assuming roughly three weeks remain, sequence it like this:
Days 1–2 — Establish scope. Run the codebase audit. Trigger the data export. Confirm the deprecation page still reads as expected.
Days 3–4 — Decide the path. If volume justifies it, contact OpenAI sales about dedicated capacity. In parallel, shortlist two migration targets and get API keys.
Days 5–9 — Prototype. Port your ten most-used prompts to both candidates. Compare output honestly against real Sora results, not against memory.
Days 10–14 — Build the adapter. Write an abstraction layer rather than a direct swap, so the next vendor change is a config edit. Handle the async pattern differences — polling intervals and job lifecycles vary.
Days 15–17 — Ship behind a flag. Run both paths in parallel while Sora still works. This is your only window to compare live, and it closes permanently.
Days 18–20 — Cut over and clean up. Flip the flag. Purge stored sora-2 strings from your database. Remove the dead client code.
🚨 Verify the date yourself before you plan around it. Deprecation dates can move, and OpenAI has extended shutdowns before. Load developers.openai.com/api/docs/deprecations and confirm the Videos API row still reads as described here. It takes fifteen seconds and it is the only source that counts.
Frequently Asked Questions
When exactly does the Sora API shut down?
September 24, 2026. OpenAI notified affected developers on March 24, 2026 that the Videos API and Sora 2 model aliases and snapshots would be removed on that date. This is separate from the consumer app shutdown, which already happened on April 26, 2026.
Is there an OpenAI replacement for the Sora API?
No. The recommended-replacement column on OpenAI's deprecations page is empty for the Videos API and Sora 2 row, and no successor video model has been announced. Every migration path leads to a different vendor.
Does the Sora API still work right now?
Yes, until the shutdown date. The consumer app closing in April did not affect API availability — OpenAI structured this as a two-stage shutdown. Continued operation today is expected and is not evidence the deadline changed.
Can I keep access after the shutdown date?
Possibly, in limited circumstances. OpenAI's deprecations page states that in some cases developers may be able to provision dedicated capacity for continued access after a model's shutdown date, and directs interested teams to its sales department. This is discretionary and realistically aimed at enterprise volume, but it costs nothing to ask.
What is the best Sora API alternative?
It depends on your use case. Google Veo 3.1 is most often cited as the closest feature parity for general text-to-video with native audio. Kling 3.0 is typically the cheapest credible option at volume. Runway Gen-4.5 suits editing-led workflows. None is a drop-in replacement — prompt syntax does not transfer cleanly.
How do I export my Sora content before it is deleted?
Go to sora.chatgpt.com/sunset and click Export; you will be emailed when the archive is ready. OpenAI has said data will be permanently deleted after discontinuation and after any final export window, but has not committed to offering one — so export now rather than relying on a grace period.
Can I use a third-party platform to keep Sora access longer?
Not reliably. Reseller access ends whenever the reseller decides — at least one major aggregator sunset its sora-2 route in late April 2026, five months before OpenAI's own date, and now returns HTTP 410 for those requests. Routing through a proxy replaces a known deadline with an unknown one.
The Takeaway
This is not a routine model bump. The absence of a successor in OpenAI's replacement column is the whole story — it means you are not upgrading, you are changing vendors, with everything that implies for prompt libraries, output expectations and integration surface.
Three things today, in this order: run the grep across every repo and every database column that stores a model ID as a string. Trigger the export, because the grace period is conditional and may not exist. And if your volume is meaningful, send one email to OpenAI sales asking about dedicated capacity before you assume a hard cutover is your only option.
Then build the adapter layer rather than a direct swap. Sora's shutdown is a useful lesson about single-vendor dependency in a fast-moving category, and the cheapest time to act on that lesson is while you are already in the code.
Dates, affected model IDs and the dedicated-capacity provision in this article are drawn from OpenAI's official API deprecations documentation and OpenAI Help Center guidance on the Sora discontinuation. Deprecation schedules can change — verify the current state of the Videos API row on OpenAI's deprecations page before planning a cutover around any date given here.
🚀 Stay Connected With Simple AI Tools
API changes, AI tooling and automation guides — verified, not hyped.
👇 💬 Drop your comment below and let us know your thoughts! ✨
Comments
Post a Comment