Skip to main content
The Sora 2 video API is retired: sora-2 and sora-2-pro can no longer be used for new requests on LaoZhang API. Stop sending Sora 2 jobs, save any finished videos you still need, and move new video work to Wan 2.7 or Seedance 2.0.
  • Published: September 24, 2026
  • Last verified: October 31, 2026
  • Status: Retired

​
What changed

The retirement applies to Sora 2 requests. Seedance 2.0 still uses GET /v1/videos/{id}/content to download Seedance results, so you’ll see that path in the Seedance guide.

​
Who is affected

You are affected if you:
  • send sora-2 or sora-2-pro requests to /v1/videos, including through the OpenAI SDK videos resource with the LaoZhang API base URL;
  • created tokens in the OpenAI Official or OpenAI Official Enterprise group only for Sora 2;
  • have finished Sora 2 videos that you haven’t copied to your own storage yet.
You are not affected if you only use Wan 2.7, Seedance 2.0, or image models. Image requests that share the same token groups, such as official-forward gpt-image-2 and the GPT Image 2.5 Flare and Sunburst models, are not part of this retirement. See the GPT Image 2.5 guide and the GPT Image 2 guide for those routes.

​
What to do now

  1. Stop creating Sora 2 jobs. Remove sora-2 and sora-2-pro from request code, retry logic, schedulers, and job queues. New requests won’t produce video, so retries won’t help.
  2. Save finished videos. Download every completed Sora 2 video you still need and keep it in your own storage. OpenAI download links are short-lived, and the Videos API has been removed upstream, so a video you haven’t saved may no longer be retrievable.
  3. Move to Wan 2.7 or Seedance 2.0. Compare the two options below, map your request fields, and run a test task before you switch production traffic. To change as little code as possible, move to Muse, which keeps the same endpoints; see Smallest change: move to Muse.
  4. Clean up your tokens. In token management, delete tokens you created only for Sora 2. Keep OpenAI Official and OpenAI Official Enterprise tokens that also serve official-forward image models. Create a new token in the Wan group for Wan 2.7, or in the SeeDance2 group for Seedance 2.0.

​
Smallest change: move to Muse

Muse, Meta’s video model, launched on October 30, 2026 and keeps the OpenAI video endpoints that Sora 2 used, so your create, poll, and download code can stay:
  • Create with POST /v1/videos, poll with GET /v1/videos/{id}, and download with GET /v1/videos/{id}/content.
  • Multipart first-frame uploads still use the input_reference field, and size still accepts 1280x720 and 720x1280.
  • Status values are the same: queued, in_progress, completed, failed.
Only three things change:
  • Set model to muse-video. For clips of 20 to 60 seconds, use muse-video-20s to muse-video-60s.
  • Set seconds to "5", "8", or "10". Sora 2’s 4- and 12-second lengths aren’t accepted.
  • Use a token in the default group with the Usage first billing mode.
Muse bills $0.30 per video for 5–10 seconds and outputs at the 720p tier with audio. For prices next to Wan, Seedance, and MiniMax H3, see the video generation API comparison.

​
Choose a replacement

Both replacements are asynchronous: you create a task, save the task ID, poll until it finishes, and then download an MP4.
  • Choose Wan 2.7 when you want to edit an existing clip or animate an image in sync with an audio track.
  • Choose Seedance 2.0 when you want to combine image, video, and audio references in one request, or to keep a real person’s likeness through the real-person and virtual human asset API.
If you’re not sure, send the same prompt to both and compare the results and the charges. Each guide’s billing section gives the USD prices, the conversion from the official list price, and per-task estimates, based on public pricing in effect since October 14, 2026. The charge for each task appears in call logs. Seedance 2.0 writes two rows per task, a pre-charge and a settlement, and the task’s cost is their sum.

​
Endpoints

​
Map your Sora 2 request

Accepted durations and resolutions differ by model. Check the parameter tables in the Wan 2.7 guide and the Seedance 2.0 guide before you port values from Sora 2.

​
Test before you switch traffic

Create one short task with the new model and save its task ID. Poll until it succeeds, download the MP4, and check that it plays. Then confirm the model, token group, and charge in call logs before you move production traffic.

​
Frequently asked questions

​
Can I still call sora-2 or sora-2-pro?

No. LaoZhang API retired the route on July 1, 2026, and OpenAI removed both models and the Videos API on September 24, 2026. New requests won’t produce video. Use Wan 2.7 or Seedance 2.0 instead.

​
Why do sora-2 and sora-2-pro still appear in the price list?

The public price list may show them for a while after the retirement. A listing doesn’t make a model usable: OpenAI has removed both models from its API. To check what a token can call, follow how to confirm model availability.

​
Can I still download videos I generated earlier?

Try now, and keep your own copy of anything you need. OpenAI download links are short-lived, and with the Videos API removed upstream, a result you haven’t saved may no longer be available.

​
Can I keep my OpenAI SDK videos code and change only the model name?

No. Wan 2.7 and Seedance 2.0 don’t create tasks through /v1/videos. Replace the client.videos calls with requests to the Wan or Seedance task endpoints listed above. Both guides include cURL and Python examples.

​
Do I need a new token?

Yes. Wan 2.7 needs a token in the Wan group, and Seedance 2.0 needs a token in the SeeDance2 group. Tokens in OpenAI Official or OpenAI Official Enterprise can’t call these models.

​
Archived legacy route docs

These pages document the earlier Sora 2 legacy route: the sora_video2 models, the async task queue, and character creation. They’re kept for existing integrations only, and their examples no longer work.