Model comparison

Wan 3.0 vs Seedance 2.5: 30s Each, 20 vs 50 Refs

Both ship 30 seconds in a single generation pass. Wan 3.0 wins on document input and per-second pricing; Seedance 2.5 wins on reference count (up to 50) and is generally available with published per-second pricing.

Decision snapshot

Decision factorWan 3.0Seedance 2.5
Single-pass duration ceiling30 seconds30 seconds
Reference count per requestUp to 20Up to 50
Document input (PDF, PPT)YesNo
Audio in single passYesYes
Published per-second pricingYesYes
Open weightsNoNo
Public API rank on Artificial Analysis1,240 text-to-video1,180 image-to-video

Where each model wins

  • Document-to-video and slide-deck-to-video — only Wan 3.0 reads documents.
  • Heavy reference mixing (50+ inputs) — only Seedance 2.5.
  • Strict open-weight self-host — neither qualifies; pick Kling 2.5.
  • Cheapest per-second at 720p — Wan 3.0 at $0.10/s.

Cost on the same workload

A 30-second 720p clip costs $3.00 on Wan 3.0. Seedance 2.5 publishes a similar per-second rate at the time of writing but is staged-access; verify the current rate before committing production.

Ready to test the workflow?

Create account & add credits

Workload recommendation

WorkloadRecommendation
PDF, slides, webpage URL → videoWan 3.0 — only it accepts documents.
Long reference-driven scenes (20-50 refs)Seedance 2.5 — more reference room.
30-second explainer clipBoth — choose by reference count need.
Audio-in-pass narrationBoth — both compose audio in one pass.
Cheapest per-second at 720pWan 3.0 — published at $0.10/s.

Benchmark rules

  • Compare only the same harness version.
  • Vendor-reported and independently measured results are separate evidence classes.
  • Record clip duration, resolution, and audio settings alongside any score.
  • Seedance 2.5 is in staged access; verify current availability.

Frequently asked questions

Can I try Wan 3.0 before integrating it?

Use the OneInfer Wan 3.0 launcher to open a prepared prompt in the public playground. Availability is checked against the current model catalog.

How should I treat benchmark claims?

Check the provenance label and harness version. Vendor-reported and independently verified results are deliberately shown as different evidence classes.

Put Wan 3.0 to work

Fund a controlled evaluation, send a reference frame or document, and measure quality and cost on your own workload.