Which Wan models are on varg?

3: Wan 3.0 (from $0.10 per run), Wan 2.5 (from $0.25 per run) and Wan 2.5 Preview (from $0.25 per run). Each has its own page with the parameters and a call to copy.

Which Wan models take an image?

All 3: Wan 3.0, Wan 2.5 and Wan 2.5 Preview. The catalog lists image to video for them; pass the image as a URL in "files".

What is the longest clip the Wan models make?

30 seconds, on Wan 3.0, by the models' input schemas. Then Wan 2.5 (10 s) and Wan 2.5 Preview (10 s).

How long do Wan requests take?

Typically from about 2 min (Wan 2.5) to about 3 min (Wan 3.0), by the catalog's estimates for each model's fastest route. The API answers at once with a job id, so nothing waits on an open connection while it runs.

How do I call the Wan API?

Send POST https://api.varg.ai/v2/tools/video/call with the header "Authorization: Bearer $VARG_API_KEY" and a JSON body that includes "model", for example "model": "wan_3". The call answers 202 with a job right away. Poll GET https://api.varg.ai/v2/jobs/{id} until "status" is "completed", then read output.outputs[0].url, or pass options.webhook_url to be called back.

Can I use Wan from Claude, ChatGPT or Cursor?

Yes. Add the varg MCP server (https://mcp.varg.ai/mcp) to Claude, ChatGPT, Cursor or any other MCP client and sign in with varg, with no API key to paste. The agent runs a model with call_tool, passing tool "video" and the model id, for example "model": "wan_3".

3 models · video

Wan API

Wan is a family of AI video models made by Alibaba. varg runs 3 of them on one API key: Wan 3.0, Wan 2.5 and Wan 2.5 Preview. Between them they cover text to video, image to video and video to video; they return video.

Price: From $0.10 per run Wan 3.0
Made by: Alibaba
Flagship id: wan_3

Every Wan model runs on one varg API key and one credit balance, the same as every other model here. You pay the provider's price, with no markup.

POST /v2/tools/video/call

curl -X POST https://api.varg.ai/v2/tools/video/call \
  -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "wan_3",
  "prompt": "A red paper lantern drifts up through a rainy night market in Hong Kong, neon signs reflecting in wet streets, steam rising from food stalls. The camera follows the lantern as it rises above the crowd and reveals the glowing city skyline. Sounds of rain, sizzling woks and distant chatter.",
  "aspect_ratio": "16:9",
  "resolution": "720p",
  "audio": true,
  "duration": 5
}'

Try Wan 3.0 in the Playground Get an API key

Wan models and prices

Wan on varg starts from $0.10 per run (Wan 3.0). Prices are in US dollars and are the provider's own, with no markup. A "from" price is the cheapest request the catalog states for that model; each model's page has every rate and input. The descriptions are the catalog's own, shortened.

Wan models on varg
Model	Catalog description	Takes	Price	Typical time
Wan 3.0`wan_3`	Alibaba Wan 3.0. Top-ranked text-to-video — up to 30s at 1080p with native audio, first/last-frame animation, and references (≤10 images, ≤5 videos…	Text to video, Image to video, Video to video	From $0.10 per runfrom $0.05/s	~3 min
Wan 2.5`wan_2_5`	Alibaba Wan 2.5. Strong text-to-video at competitive price.	Text to video, Image to video	From $0.25 per runfrom $0.05/s	~2 min
Wan 2.5 Preview`wan_2_5_preview`	Preview variant with lip-sync support.	Text to video, Image to video	From $0.25 per runfrom $0.05/s	~2 min

Call the Wan API

The call below runs Wan 3.0. For another Wan model, put its id from the table in "model"; the rest of the request keeps its shape.

REST

POST https://api.varg.ai/v2/tools/video/call with Authorization: Bearer $VARG_API_KEY. It answers 202 with a job right away; poll GET /v2/jobs/{id} until its status is completed, then read output.outputs[0].url.

POST /v2/tools/video/call

curl -X POST https://api.varg.ai/v2/tools/video/call \
  -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "wan_3",
  "prompt": "A red paper lantern drifts up through a rainy night market in Hong Kong, neon signs reflecting in wet streets, steam rising from food stalls. The camera follows the lantern as it rises above the crowd and reveals the glowing city skyline. Sounds of rain, sizzling woks and distant chatter.",
  "aspect_ratio": "16:9",
  "resolution": "720p",
  "audio": true,
  "duration": 5
}'

From Claude, ChatGPT or Cursor (MCP)

The varg MCP server, https://mcp.varg.ai/mcp, runs Wan 3.0 with call_tool. Sign in with varg, no API key to paste. How to connect it

Claude Code

claude mcp add --transport http varg https://mcp.varg.ai/mcp

call_tool

{
  "tool": "video",
  "input": {
    "model": "wan_3",
    "prompt": "A red paper lantern drifts up through a rainy night market in Hong Kong, neon signs reflecting in wet streets, steam rising from food stalls. The camera follows the lantern as it rises above the crowd and reveals the glowing city skyline. Sounds of rain, sizzling woks and distant chatter.",
    "aspect_ratio": "16:9",
    "resolution": "720p",
    "audio": true,
    "duration": 5
  },
  "wait": true
}

From a terminal (CLI)

vargai is the varg command line: every tool of the MCP server is a command. Install and log in

vargai

curl -fsSL https://varg.ai/install | sh
vargai call_tool --tool video '{"model":"wan_3","prompt":"A red paper lantern drifts up through a rainy night market in Hong Kong, neon signs reflecting in wet streets, steam rising from food stalls. The camera follows the lantern as it rises above the crowd and reveals the glowing city skyline. Sounds of rain, sizzling woks and distant chatter.","aspect_ratio":"16:9","resolution":"720p","audio":true,"duration":5}'

How billing works

Every model is billed at its provider's price with no markup, from one credit balance: 1 credit is $0.01. varg's fee is on the top-up, not on the run: 5.5% of the credits you buy ($0.80 at the least), or 8% for a team workspace. Pricing

Wan compared

Price, limits and inputs side by side, from the same catalog as this page.

How to use Wan

Step-by-step pages on varg how to.

Wan API: questions

How much does the Wan API cost on varg?: The lowest price the catalog states in the Wan family is from $0.10 per run (from $0.05/s), on Wan 3.0. Every model is billed at its provider's price with no markup, from one credit balance: 1 credit is $0.01. For the exact price of a request before you run it, send the same body to POST https://api.varg.ai/v2/estimate.
Which Wan models are on varg?: 3: Wan 3.0 (from $0.10 per run), Wan 2.5 (from $0.25 per run) and Wan 2.5 Preview (from $0.25 per run). Each has its own page with the parameters and a call to copy.
Which Wan models take an image?: All 3: Wan 3.0, Wan 2.5 and Wan 2.5 Preview. The catalog lists image to video for them; pass the image as a URL in "files".
What is the longest clip the Wan models make?: 30 seconds, on Wan 3.0, by the models' input schemas. Then Wan 2.5 (10 s) and Wan 2.5 Preview (10 s).
How long do Wan requests take?: Typically from about 2 min (Wan 2.5) to about 3 min (Wan 3.0), by the catalog's estimates for each model's fastest route. The API answers at once with a job id, so nothing waits on an open connection while it runs.
How do I call the Wan API?: Send POST https://api.varg.ai/v2/tools/video/call with the header "Authorization: Bearer $VARG_API_KEY" and a JSON body that includes "model", for example "model": "wan_3". The call answers 202 with a job right away. Poll GET https://api.varg.ai/v2/jobs/{id} until "status" is "completed", then read output.outputs[0].url, or pass options.webhook_url to be called back.
Can I use Wan from Claude, ChatGPT or Cursor?: Yes. Add the varg MCP server (https://mcp.varg.ai/mcp) to Claude, ChatGPT, Cursor or any other MCP client and sign in with varg, with no API key to paste. The agent runs a model with call_tool, passing tool "video" and the model id, for example "model": "wan_3".

More model APIs

Same API key, same request shape: change the model string to switch.

All models

For AI agents: each model above has its full spec as Markdown at varg.ai/models/wan_3.md (swap the id), and every model is listed in varg.ai/models/llms.txt.