OpenAI Video API Shutdown: What to Replace After Sora 2

OpenAI's Videos API and Sora 2 models have reached their published removal date. OpenAI lists no replacement video model or endpoint. A code example that remains visible in the old video guide is not a basis for a new production integration.

If your application used Sora, identify every video job path, preserve outputs you need, and move generation to another provider. This guide maps the affected calls and four video routes currently published on Modellix. Those routes are alternatives to evaluate, not drop-in replacements for Sora.

Sora 2 shutdown explained: dates and affected IDs

OpenAI notified developers on March 24, 2026 that it would remove the entire Videos API and these model IDs on September 24:

• sora-2 and sora-2-pro
• sora-2-2025-10-06, sora-2-2025-12-08, and sora-2-pro-2025-10-06

The deprecations table leaves the replacement column empty. Pinning a dated snapshot does not avoid the shutdown. The separate Sora web and app experiences ended on April 26, 2026. OpenAI's consumer content export at sora.chatgpt.com/sunset is for Sora content; it is not presented as an archive for API video jobs.

What existing Sora integrations need to replace

OpenAI published the removal date, not a universal HTTP error code or a guaranteed retrieval window for old jobs. Treat failed Sora calls as a migration signal rather than automatically retrying them. If an asset exists only in an OpenAI video job, check your account for any remaining retrieval access immediately.

4 live video routes compared by input and clip length

Each route below had a responding model page and public api_schema on September 24, 2026. That confirms the routes are published; it is not a claim that we completed a paid generation on each one. Modellix publishes this article and sells access to these models. You can also integrate with the underlying providers directly.

Check each model page for current pricing and the exact schema before submitting a paid job. A route that fits Sora's prompt-only flow may still differ in duration, resolution, content policy, output format, and cost. For a wider comparison, see our video generation API guide.

Replace the job lifecycle, not just the URL

For a first text-to-video migration test, the current Veo 3.1 Fast API reference publishes this request shape:

curl -sS -X POST "https://api.modellix.ai/api/v1/google/veo-3.1-fast-t2v" \
  -H "Authorization: Bearer $MODELLIX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A red kite rises above a grassy park at golden hour",
    "aspectRatio": "16:9",
    "duration": "8",
    "resolution": "720p"
  }'

The response returns data.task_id and a data.get_result.url. Poll that URL, or use Modellix's own webhook events; OpenAI's video.completed handler will not receive these tasks. The REST API guide documents the task response, webhook headers, and result retrieval. Modellix retains generated results for seven days, so copy the output file to your own storage promptly.

Before switching production traffic, test representative prompts and, for image-to-video jobs, the same input image. Confirm the allowed duration and resolution, inspect the generated file, and compare the billed cost. Keep your application's video job ID separate from the provider's task ID so future model changes do not strand your records.

Can you still use an OpenAI API for video?

OpenAI's published September 24 removal covers its Videos API and every listed Sora 2 alias and snapshot. Its deprecations page names no successor, and it does not identify the Responses API as a video migration path. OpenAI's other APIs are separate products; a working image API does not keep Sora video generation available.

If you need video now, select a current video provider and adapt the request, task status, output storage, and billing checks. The four Modellix routes above give you concrete starting points; the Veo 3.1 API guide goes deeper on the submit, poll, and retrieve flow.