Wan 3.0 launched on August 24, 2026 with 30-second clips and per-second billing from $0.05 to $0.20. This article breaks down Standard and Prime rates, real clip costs, the Alibaba Model Studio API setup, the official docs and the free ways to test Wan 3 before paying.
Alibaba shipped Wan 3.0 on August 24, 2026, and the first question every builder asked was simple: what does a clip cost, and how do I get in? The short answer is $0.05, $0.10 and $0.20 per generated second at 480p, 720p and 1080p on the Singapore endpoint of Alibaba Cloud Model Studio, plus a faster Prime tier at $0.068, $0.14 and $0.28. A 30-second 1080p clip lands at $6. Access runs through a region-specific API credential created in the Model Studio console, and the official video pages list no standing free tier. Below you get the rates, the math, the exact request flow, the docs worth bookmarking, and a free route through Wan 3 on PicassoIA so you can test prompts before spending a cent.
Alibaba bills by output second, and the rate depends on resolution and region. The price sheet splits into two models: the Standard model wan3.0-video and the faster wan3.0-video-prime. Both take text, images, video, audio, documents and web links as input and return one MP4 with native audio.
| Resolution | Singapore (USD per second) | Beijing, Tokyo, Frankfurt, Virginia, Hong Kong |
|---|---|---|
| 480p | $0.05 | $0.041256 |
| 720p | $0.10 | $0.082513 |
| 1080p | $0.20 | $0.165025 |
Singapore carries the highest listed rate. The other five regions run about 17 percent cheaper, but your credential only works in the region where you created it, so pick the region before you build anything around it.
Wan 3 Prime is the accelerated version. Alibaba describes it as offering significantly faster generation while keeping output quality high, and it costs roughly 36 to 40 percent more per second in Singapore.
| Resolution | Singapore (USD per second) | Other five regions |
|---|---|---|
| 480p | $0.068 | $0.0636 |
| 720p | $0.14 | $0.1272 |
| 1080p | $0.28 | $0.2544 |
How much faster is it? PicassoIA's own sample renders give a rough picture: a 5-second 1080p clip on Wan 3 took about 322 seconds, while a 5-second 1080p sample on Prime finished in about 102 seconds, and a 480p Prime draft took 36 seconds. The prompts differed, so treat those numbers as a ballpark, not a benchmark. A sensible split is Standard for overnight batches where nobody is waiting, and Prime for client reviews and live iteration where every minute of waiting costs more than the price gap.
After the August 24 launch, Alibaba ran a 30 percent discount on the Standard tier, which pulled Singapore 720p down to about $0.07 per second. Reports put the end date at September 23 or 24, 2026. The Model Studio pages, last updated on September 28, 2026, list the full rates above, so budget with those numbers. Third-party host fal reportedly lists the same $0.05, $0.10 and $0.20 rates for Wan 3, which means no hidden markup for going through a reseller.
💡 Budget tip: Launch reports said Wan 3.0 undercuts Google's Veo 3.1 standard tier, reported at $0.40 per second. At 1080p, Wan 3.0 Standard costs half of that.
A per-second price hides how quickly clips add up. Duration is the biggest lever you control, because Wan 3.0 accepts anything from 2 to 30 seconds per generation.
| Length | 480p | 720p | 1080p | 1080p Prime |
|---|---|---|---|---|
| 5 seconds | $0.25 | $0.50 | $1.00 | $1.40 |
| 10 seconds | $0.50 | $1.00 | $2.00 | $2.80 |
| 15 seconds | $0.75 | $1.50 | $3.00 | $4.20 |
| 30 seconds | $1.50 | $3.00 | $6.00 | $8.40 |
The price sheets I checked list one rate per resolution and no separate audio line, so the audio switch is a creative choice about dialogue, music and sound effects rather than a listed discount. All figures use Singapore rates. A full 30-second 1080p take at $6 is a fair price for a single pass that previously needed two or three stitched clips, since Wan 2.7 topped out at 15 seconds.
| Scenario | Volume | Math | Monthly cost |
|---|---|---|---|
| Solo creator | 60 clips, 5 s, 720p | 300 s × $0.10 | $30 |
| Small shop | 100 clips, 10 s, 720p | 1,000 s × $0.10 | $100 |
| Agency | 40 clips, 30 s, 1080p | 1,200 s × $0.20 | $240 |
Resolution is the cheapest dial to turn. A 480p render costs a quarter of a 1080p render at the same length. Say you need 20 finished 10-second clips and each one takes five attempts:
That is a $110 difference on one batch. Draft renders won't match the final frame for frame, so keep the same prompt and seed when you move up.
Model Studio issues one credential per region, and each region has its own endpoint, credential list and model list. They cannot be mixed.
💡 Access note: An early third-party write-up from August 10 described Wan 3.0 as a preview with invitation-only pricing. The model page updated on September 28 mentions a playground and no invitation restriction. If your account still sees a gate, check the Model Studio console for application status.
Open the API credential page in the Model Studio console (under Settings, with your region selected in the upper right corner) and click Create. Leave the workspace on Default, add an optional description of up to 200 characters, and set permissions to All or Custom. The plaintext value appears only once, so copy it into a secret manager or an environment variable immediately.
sk-ws and cannot be viewed again.
The API is asynchronous: you submit a task, receive a task ID, then poll until the video is ready.
Each region has its own host pattern, with your workspace ID in front:
ap-southeast-1.maas.aliyuncs.comcn-beijing.maas.aliyuncs.comus-east-1.maas.aliyuncs.comap-northeast-1.maas.aliyuncs.comeu-central-1.maas.aliyuncs.comcn-hongkong.maas.aliyuncs.comA text-to-video call to Singapore looks like this:
curl -X POST "https://$WORKSPACE_ID.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis" \
-H "Authorization: Bearer $MODELSTUDIO_TOKEN" \
-H "Content-Type: application/json" \
-H "X-DashScope-Async: enable" \
-d '{
"model": "wan3.0-video",
"input": { "prompt": "A fishing boat leaves a harbor at sunrise, slow aerial pull-back, gulls crossing the frame" },
"parameters": { "resolution": "480P", "duration": 10, "ratio": "16:9", "audio": true }
}'
The X-DashScope-Async: enable header is mandatory. Parameters you can set include resolution (480P, 720P or 1080P, default 1080P), duration (2 to 30, or -1 for smart duration), ratio (adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16), audio and prompt_extend.
The first response returns a task_id with status PENDING. Query it with a GET request to /api/v1/tasks/{task_id} on the same host. The status moves from PENDING to RUNNING to SUCCEEDED or FAILED. Poll about every 15 seconds.
💡 Download fast: Both the task ID and the video URL stay valid for only 24 hours. Save the MP4 to your own storage as soon as the task succeeds.
| Page | What you find there |
|---|---|
| Wan3.0-Video model information | Standard pricing by region, rate limit, supported inputs |
| Wan3.0-Video-Prime model information | Prime pricing by region |
| Wan3.0 Video Generation API Reference | Endpoints, headers, request body, task polling |
| Alibaba Cloud Community launch post | Capabilities and launch context |
Each of those pages shows a "last updated" date, and prices have already moved once since launch, so check the date before you quote a number to a client. Plan around these limits from the docs: 300 requests per minute across all regions, prompts up to 20,000 characters, up to 10 reference images, 5 reference videos (15 seconds total, 100 MB each), 5 reference audio files (15 seconds total, 15 MB each), and one document plus one web link. Output is MP4 at 30 fps.
Model Studio's new-user free quota is 1 million tokens valid for 90 days, offered only in Singapore, and it applies to Qwen text models. The Wan 3.0 pages list no free quota. One aggregator reported 30 free seconds of video for new users, but that figure does not appear in the official docs, so check your console's free quota page before relying on it. Consumer routes are thin too: launch reports described the wan.video site as members-only, and one access write-up said the Qwen app and official Wan site offer only small trial allocations.
Treat free Wan 3.0 API access as unconfirmed and plan on paying the metered rate.
Wan 3 and Wan 3 Prime are both listed on Picasso IA as free to try online, with no credential, workspace or region to set up. That makes them the cheapest way to test prompts before moving a workflow to the paid API. If you need unlimited volume for testing, two free options also exist: Seedance 2.5 Lite, listed as free and unlimited for clips up to 10 seconds, and PicassoIA Video, listed as a free unlimited generator from text or image.
Here is the fast path from an idea to a clip with Wan 3:
Automatic prompt expansion is on by default. It improves short prompts but adds latency, so switch it off when your prompt is already detailed.
A ceramic mug rotates slowly on a linen cloth, soft studio light from the left, slow dolly-in, gentle steam rising, shallow depth of field, natural color
That prompt names one subject, one motion and one camera move, which is what keeps clips steady.
| Model | Best for |
|---|---|
| Wan 3 | Single-pass clips up to 30 seconds at up to 1080p |
| Wan 3 Prime | The same specs with faster renders |
| Wan 2.7 T2V | Text to 1080p video, clips up to 15 seconds |
| Wan 2.7 I2V | Animating a single photo |
| Wan 2.7 R2V | Animating a reference subject |
| Wan 2.6 T2V | HD text to video at an older generation |
One difference matters for licensing. Alibaba's published open weights stop at Wan 2.2 under Apache 2.0, while Wan 3.0 is closed weights. You cannot self-host it, so the API or a hosted platform is the only route.
Alibaba's docs say Wan 3.0 can edit existing video and extend it forward, backward or both ways through the API. On PicassoIA you can split that work across dedicated tools: Wan 2.7 Videoedit for text-driven edits, Lucy Edit 2 for restyling, P Video Edit for prompt-based changes, and Video Increase Resolution to upscale a 480p draft you decided to keep.
Before you commit real budget to the paid API, run through this short checklist:
duration never turns a $1 test into a $6 one.Pricing pages are useful, but a real render tells you more than any table. Open Wan 3 on Picasso IA, paste the mug prompt above, and generate a 480p draft. Then run the same prompt on Wan 3 Prime and compare speed. Once you know which prompts hold up, move the winners to the paid Alibaba API with a clear per-clip budget, or keep everything on Picasso IA and browse every model at picassoia.com/en/all-models. Experiment, change one variable at a time, and let your own footage set the budget.
