Sora files lived 1 hour: storing Sume job results

OpenAI's docs say Sora 2 and the Videos API shut down on Sept 24, 2026, that both models produced 16 and 20 second generations, that videos could be downloaded for up to 1 hour after completion, and that no direct replacement API exists. Sume results are fetched through the job's unsigned_urls or the content endpoint; the docs state no retention period, so copy the file to your own storage as soon as the job completes.

What the Sora rules teach

A one-hour download window forces a workflow where the file is copied right after completion. Any pipeline that stored only the provider's URL would have lost the asset an hour later. The lesson applies to every provider: the job result is a pointer, and you own the copy.

Downloading on Sume

Poll GET /v1/videos/{jobId} until status is completed, then read unsigned_urls. You can also call the content endpoint directly:

curl "https://api.sume.com/v1/videos/{jobId}/content?index=0" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  --output video.mp4

Where results live

The Sume docs show a successful poll response with unsigned_urls and a usage.cost field. For jobs read through /v1/jobs/{id}/result, artifacts carry a Sume-hosted URL on media.sume.com. Raw provider URLs are not public outputs, so you never depend on a provider link.

A storage routine

Treat the file as yours the moment it exists:

• On completion, download the file and write it to your own bucket
• Record the job id next to the stored object key
• Do not resubmit a paid job to recover a file you could have stored
• Keep the model id and prompt with the file so you can reproduce a variant