Sora 2 vs Wan 3.0 after the API shutdown: length, size and price

If your Sora 2 jobs were plain text-to-video clips, wan-3.0 is the closest swap on Sume: it takes 2 to 30 seconds, renders at 480p, 720p or 1080p, and accepts reference images, videos and audio. The Sora 2 models and the Videos API were shut down on September 24, 2026, so the question is no longer which is better but which limits your old calls actually used.

What shut down, and what OpenAI lists instead

According to OpenAI's deprecations page, the shutdown covers the Videos API and the aliases sora-2, sora-2-pro, sora-2-2025-10-06, sora-2-2025-12-08 and sora-2-pro-2025-10-06, with no replacement listed. That is why the table below puts the Sora limits (from OpenAI's own guide) next to the Wan limits Sume documents.

The practical effect for a Sora integration is that every model alias you hard-coded stopped working on the same day. A search of your codebase for sora-2 and sora-2-pro is the fastest way to find each place that needs a new id, and the cheapest first fix is a single config value for the model name.

Limits side by side

What a clip costs

Sume bills list price times 1.25, reserved when you submit and rounded up to the cent, as described on the generation admission page. A 10 second 720p Wan clip is therefore 10 x $0.125 = $1.25, and the same clip at 1080p is $2.50. A failed job releases or refunds the reservation.

Quality tiers are the other thing to rethink. If you routed hero clips to sora-2-pro and drafts to sora-2, the closest habit on Sume is one id with two resolutions, not two ids. That keeps the prompt identical between a cheap look and the final render, so what you approve at 480p is what you pay to see at 1080p.

Porting the two-tier habit

Sora's two tiers split quality by model name. On Sume you keep one model id and move the cost with the resolution field instead, so a draft pass at 480p costs a quarter of the 1080p final for the same prompt.

Before you cut over, render the same three prompts on wan-3.0 and compare them with your old Sora outputs at the same length. Judge motion, text legibility and how the model handles the references you actually send, then decide whether 720p is enough for your channel. Most feeds do not need more.

• Draft at 480p, $0.0625 per second, then re-run only the keepers at 1080p.
• Use durations up to 30 seconds when you used to stitch several 16 or 20 second clips.
• Keep prompts as they are; port the request shape, not the wording.

Request changes

The request is one POST to /v1/videos with model, prompt, duration, resolution and aspect_ratio. Aspect ratios for Wan are auto, adaptive, 16:9, 4:3, 1:1, 3:4 and 9:16. If your old call sent size or seed, note the Sume route rejects both with 400 unsupported_parameter, so drop both rather than translating them. The full field list is in the video generation docs.