Wan 3 on ModelRush gives you more creative control, when your tired of "this prompt violates our guidelines"!
Text, a first and last frame, or reference images, video and audio, at any whole-second length from 2 to 30 seconds.
A 10-second 720P clip costs $0.90, against $0.83 on Alibaba Cloud's Virginia region and $1.00 on fal. Here is what you can control, what each provider charges, and which one fits your workload.
What you can control on ModelRush
- Where the shot starts and ends. Image to video takes a first frame and an optional last frame, so you set the opening and closing picture.
- What the shot is guided by. Reference to video takes up to 10 images, 5 videos and 5 audio files. Keyframes and references are separate modes and cannot be combined.
- Length, shape and sound. Any whole second from 2 to 30, at 480P, 720P or 1080P, with your choice of aspect ratio and audio on or off.
- What it costs before you run it. The playground shows the price for your exact settings, the model's average return time and its failure rate. Failed jobs are not billed.
Which one should you pick
| If you need | Pick | Why |
|---|---|---|
| Frames, references, length and audio set in one OpenAI-style request, under fal's price | ModelRush | $0.09/s at 720P, 10% under fal, no charge for failed jobs |
| Lowest published per-resolution rate and a direct contract with the model owner | Alibaba Cloud, US Virginia | $0.0825/s at 720P on the native DashScope API |
| Your team already runs on fal's SDK and queue | fal | $0.10/s at 720P |
| One OpenAI-compatible key and a single headline rate | Atlas Cloud | Page lists $0.04/s with no per-resolution breakdown; confirm the 720P and 1080P rate before you commit |
Price per second
All rates are per second of generated video on the standard Wan 3 tier (not Prime), read from each provider's own page: ModelRush and fal on 18 Nov 2026, Alibaba Cloud and Atlas Cloud on 19 Nov 2026.
| Provider | 480P (USD/s) | 720P (USD/s) | 1080P (USD/s) | 10 s at 720P |
|---|---|---|---|---|
| Alibaba Cloud, US Virginia (global) | 0.041256 | 0.082513 | 0.165025 | $0.83 |
ModelRush (modelrush/wan3) |
0.045 | 0.09 | 0.18 | $0.90 |
fal (alibaba/wan-3.0) |
0.05 | 0.10 | 0.20 | $1.00 |
| Alibaba Cloud, Singapore (international) | 0.05 | 0.10 | 0.20 | $1.00 |
Atlas Cloud's Wan 3 page lists a single $0.04 per second and does not say which resolution it covers, so it isn't in this table. If that rate holds at 720P it would be the lowest here. Check the price shown in your Atlas console before you compare.
A worked monthly bill
A short-drama studio renders 200 clips a month, 10 seconds each, at 720P. That is 2,000 billable seconds.
| Provider | Rate at 720P | Monthly bill |
|---|---|---|
| Alibaba Cloud, US Virginia | $0.082513/s | $165.03 |
| ModelRush | $0.09/s | $180.00 |
| fal | $0.10/s | $200.00 |
| Alibaba Cloud, Singapore | $0.10/s | $200.00 |
Retries change the real number more than the rate does. If 1 in 4 clips is re-rolled, multiply every row by 1.25. On ModelRush, failed jobs are not billed, so a render that errors out costs $0.
Specs and API shape
The model is the same everywhere: Alibaba's wan3.0-video, clips up to 30 seconds, 480P to 1080P, audio in the same pass. What differs is the wrapper.
| Provider | Endpoint style | Auth | Async pattern | Model ID |
|---|---|---|---|---|
| ModelRush | OpenAI-style POST /v1/videos/generations
|
Bearer key | Poll GET /v1/videos/generations/{id} or webhook |
modelrush/wan3 |
| Alibaba Cloud | DashScope native | Bearer key + X-DashScope-Async
|
Poll task ID | wan3.0-video |
| fal | fal queue API, per-endpoint URLs |
Key header |
Queue status + result | alibaba/wan-3.0/... |
| Atlas Cloud | OpenAI-compatible | Bearer key | Poll | alibaba/wan-3.0/... |
When to pick each
- ModelRush if you want to direct the shot: set the frames, references, length and audio in one OpenAI-style request, see the price before you run it, and pay nothing for failed jobs. It sits 10% under fal, next to image, voice and text models on the same key.
- Alibaba Cloud if price is the only line item and you can run the DashScope API and contract directly. Virginia is 8% under ModelRush at every resolution.
- fal if your team already runs on fal's SDK and queue and the extra $0.01/s at 720P is worth not switching.
- Atlas Cloud if its $0.04/s headline rate applies to the resolution you render. Confirm that first.
The request on ModelRush
curl https://api.modelrush.ai/v1/videos/generations \
-H "Authorization: Bearer $MODELRUSH_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"modelrush/wan3","prompt":"A ceramic mug on a cream background, slow dolly in","duration":10,"resolution":"720P","ratio":"16:9","audio":true}'
# Returns 202 with a generation id, status "queued" and a poll_url.
# Poll until status is "completed", then read video_url:
curl https://api.modelrush.ai/v1/videos/generations/$GENERATION_ID \
-H "Authorization: Bearer $MODELRUSH_API_KEY"
ratio and audio set the aspect ratio and the audio track. To animate an image, add first_frame with an image URL. For production, register a webhook with POST /v1/webhooks/endpoints and use polling only as a fallback.
On 18 Nov, Wan 3 averaged 4 min 21 sec per job with 0% service failures across 97 sampled calls.
Sources
- Alibaba Cloud Model Studio: wan3.0-video pricing by region
- fal: Wan 3.0 text-to-video
- Atlas Cloud: Wan 3.0 text-to-video
- ModelRush: Wan 3
Full disclosure: I work at ModelRush.

Top comments (0)