Storyboard frames flowing through an AI video production pipeline toward a final cinematic shot.
Article
Updated
4 min read
UllrAI Lab

Wan 3.0 API and Prime: Current Specs, Pricing & Release Status

Check Wan 3.0 and Prime, 2–30 second output, 1080p, regional pricing, input-video charges, and the boundary between hosted APIs and open weights.

Updated October 21, 2026: Alibaba Cloud's current catalog lists both wan3.0-video and the speed-optimized wan3.0-video-prime. The August 6 public-beta milestone remains part of the release history, but its early preview labels and price snapshot should not be used as today's universal access or billing rule.

Current Wan 3.0 API specifications

ItemOfficial catalog
Model IDswan3.0-video, wan3.0-video-prime
Duration2–30 seconds
Output480P, 720P, 1080P; 30 fps MP4
ModesText-to-video, first-frame and first/last-frame generation, all-in-one reference generation
Reference typesImages, video, audio; file and web-link context also described
Other controlsAudio toggle, smart duration, adaptive aspect ratio
Prime positioningSpeed-optimized version, according to Alibaba Cloud

The table describes the catalog, not every accepted combination. Confirm file limits, reference durations, task mode, region and account activation in the exact API you will use. Do not import Wan 2.x parameters into a 3.0 request without checking them.

Wan 3.0 pricing: input video also counts

The current international pricing page charges input-video duration plus successfully generated output duration. Failed requests are not billed under its stated rule. A previous output-only budget would therefore underestimate reference-video work.

Singapore model480P / second720P / second1080P / second
Wan 3.0 displayed list price$0.05$0.10$0.20
Wan 3.0 Prime$0.068$0.14$0.28

The normal model also carries a limited-time 30%-off notice. The table above retains the displayed list price; confirm the applied discount and expiry in your account. Tokyo, Beijing, Frankfurt, Virginia and Hong Kong have separate price entries. Do not mix currencies or regional quotes from an older article.

For a hypothetical 720P normal-model task with five input-video seconds and ten output seconds, list-price generation is 15 × $0.10 = $1.50, before any discount, free quota, retries or other production costs. Ten output seconds alone would not be the right billing duration.

Does a public catalog mean every account has access?

No. Catalog publication, API documentation and an enabled production account are separate checks. Today's catalog is more useful than repeating a beta label from August, but region, quota, authentication and model visibility still need confirmation. Submit a small controlled task before committing a production batch.

Persist the provider, region, model ID, request ID, original inputs, output settings and actual usage. Poll with bounded backoff, handle terminal errors explicitly, and copy approved output into storage you control before a temporary URL expires.

Is Wan 3.0 open source?

Hosted access does not establish an open-weight release. Do not infer 3.0 weights from the existence of Wan 2.x repositories. A self-hosting plan needs a first-party 3.0 checkpoint, inference code, a matching license and a validated runtime. This review does not establish that complete local path.

What do the launch demos show?

The official public-beta reel documents the launch context. Its selected shots are not an approval-rate benchmark.

Official launch reel. It does not disclose retries, full inputs or average production yield. Watch on YouTube

An independent continuity demonstration can help identify questions to test, but remains specific to its author's setup.

Mason AI Studio demonstration; not an Alibaba Cloud benchmark or an AniKuku test. Watch on YouTube

How to evaluate it for motion comics

Test a short reaction, a two-character exchange, hand-to-object contact, a 20–30 second action and a later matching shot. Compare normal and Prime at the same settings and retry allowance. Record end-to-end latency, rejected takes, identity errors, dialogue, human review and cost per approved second. Longer generation gives more room for storytelling and more room for errors; it is not automatically cheaper than planned short shots.

For a direct choice, see Wan 3.0 vs Seedance 2.5. Both now document 1080p, so the earlier 720p-only Seedance comparison no longer applies. Start with script-to-storyboard planning before paying for long generations.

Official sources

Reviewed October 21, 2026. Account-specific availability and billing remain subject to the live service.

Model and product names in this article are trademarks of their respective owners. AniKuku and UllrAI Lab are not affiliated with or endorsed by them.

Plan and produce your next AI video

Manage prompts, characters, storyboards, and video shots in one project, then compare models when you need to.