Wan 3.0 vs Seedance 2.5: 24 days apart, both 30 s. What differs

Seedance 2.5 launched on 2026-07-31 (ByteDance Seed) and Wan 3.0 followed about 24 days later, around 2026-08-24 (TechNode, read 2026-10-05). Both claim 30 seconds per generation. The differences are in what you can feed them and what they say about editing. On Sume both are catalog ids with a 30-second ceiling, so you can run the same prompt on each and compare.

Vendor claims side by side

The access row is the one that changes plans. The Seed page says its API is coming soon, and the Wan README describes a closed beta, so a hosted catalog is the practical way to try both side by side.

The two ids on Sume

Per the Video Router docs, seedance-2.5 accepts 4 to 30 seconds at 480p, 720p or 1080p, and wan-3.0 accepts 2 to 30 seconds. The Sume video docs also say Seedance 2.x and Wan 3.0 both accept audio and video references. The lower floor for Wan (2 seconds) matters if you want short inserts.

A fair A/B

• Same prompt, same aspect ratio, same resolution, same duration.
• Run each at least twice; neither route should be judged on one take.
• Compare on the three things you care about: subject consistency at second 25, any text in frame, and sound.
• Record the job ids so you can pull stills from the same timestamps with the video frames route.

for MODEL in seedance-2.5 wan-3.0; do
  curl -X POST https://api.sume.com/v1/video-router/generate \
    -H "Authorization: Bearer $SUME_API_KEY" \
    -H "Content-Type: application/json" \
    -H "Idempotency-Key: ab-$MODEL-001" \
    -d "{\"model\":\"$MODEL\",\"prompt\":\"A barista pours latte art, slow dolly-in, warm light\",\"resolution\":\"720p\",\"duration\":10,\"mode\":\"async\"}"
done

Reading launch-date gaps

A 24-day gap tells you who shipped first, not who is better. Both pages are vendor announcements, and both describe their own strengths. The only evidence that counts for your work is output on your prompts. Use the vendor tables to decide what to test, such as text in frame for Wan or heavy references for Seedance, and then test them.

Also keep the date in mind when you read third-party comparisons from August: they may predate updates. Re-read the vendor pages when you start a project, and note the date next to your findings.

• Use vendor tables to choose tests.
• Test on your own prompts.
• Re-read vendor pages at project start.

What to log from the A/B

Log five things per take: the model id, the prompt, the resolution, the duration and the job id. Add a one-line verdict on subject consistency, text legibility and sound. After four takes you will have a table of your own, which is worth more than any vendor table for your actual work.

Sources: Seed, Wan README, TechNode.