This is not yet a conventional model-versus-model comparison. Alibaba Cloud publicly lists Wan 2.7 reference-to-video, while it does not list an official Wan 3 or Wan 3.0 release as of August 2, 2026. Therefore, performance tables claiming a fixed winner are speculation.
What Wan 2.7 gives you today
Wan 2.7 is a concrete model-family reference you can evaluate with documented availability. Run it on your own short brief set: a product shot, a human action, a camera move, a text-heavy scene, and a reference-led continuity test. Save the input, settings, run date, output, and reviewer verdict.
How to compare a future version fairly
Keep prompts, source images, aspect ratio, duration, and review criteria fixed. Score subject consistency, instruction following, motion plausibility, first-pass acceptance, turnaround time, and cost per approved clip. Re-run the suite only after official documentation identifies the new model and scope.
Pain point: “next version” pages can make an unannounced model look available. Our added value: a test protocol that produces useful evidence now and a fair baseline later. Test the current Wan-family workflow, then compare against your own approved output—not marketing screenshots.
Sources
- Alibaba Cloud model updates — public Wan 2.7 listing.
- Wan-AI GitHub — public project materials.
- Wan-AI Hugging Face — model collection.
- Wan paper — technical context.
