Wan 3.0 Video API: One Flat Rate, No Resolution Tiers

Alibaba bills Wan 3.0 at three different rates depending on output resolution. Through Ofox it is one flat rate at any resolution, $0.05 per second of finished video. On a 1080p clip that gap is the whole story: Alibaba’s list price is 1.2 yuan per second, and the flat rate is roughly 70% below it. Everything below is read from Alibaba’s API reference and the live Ofox catalog on 4 September 2026.

Model IDs:    wan3.0-video, wan3.0-video-prime      (Alibaba Cloud)
              alibaba/wan-3.0, alibaba/wan-3.0-prime (Ofox)
Alibaba:      0.3 / 0.6 / 1.2 yuan per second at 480P / 720P / 1080P
              30% off from 2026-08-24 to 2026-09-23
Ofox:         $0.05/s flat (standard), $0.08/s flat (Prime), any resolution
Duration:     2-30 seconds; -1 selects intelligent duration
Resolutions:  480P / 720P / 1080P, default 1080p on Ofox
Aspect:       16:9, 4:3, 1:1, 3:4, 9:16, adaptive (default)
Audio:        on by default
Inputs:       text + up to 10 images + 5 videos (≤15s) + 5 audio (≤15s)
Modes:        t2v, i2v (first frame / first+last), reference, edit, continuation
Endpoint:     POST /v1/videos on Ofox; async task, poll for the result

Last updated 4 September 2026. Alibaba’s 30% discount expires 23 September 2026; the Ofox flat rate has no attached expiry.

How Much Does the Wan 3.0 API Cost?

Two different pricing models, and which one you get depends on the route. Alibaba prices the resolution. Ofox prices the second.

Alibaba’s rates are from ITHome’s launch report of the 24 August 2026 general availability, which gives the three per-second rates and the discount window of 24 August to 23 September on Bailian and the Qwen AI platform. Ofox rates are the output_video_per_second field on the live /v1/models response, read 4 September 2026.

The shape of the difference matters more than any single number. On Alibaba, moving from 480P to 1080P quadruples your bill. On Ofox, it changes nothing, and 1080p is the default resolution on the model. If your output is 1080p, the flat rate is roughly 70% under Alibaba’s list and roughly 58% under the discounted rate. If your output is 480P, Alibaba’s discounted rate is the cheaper of the two.

That is the honest framing: the flat rate wins at 1080p and loses at 480P. Which side you land on is decided by what resolution you actually ship, not by a headline.

Wan 3.0 vs Wan 2.7: The Price Halved

On Ofox, Wan 3.0 costs exactly half of Wan 2.7. Both older models bill $0.10 per second; Wan 3.0 bills $0.05.

Two models in the catalog are cheaper per second, bytedance/seedance-2.0-mini at $0.04 and bytedance/seedance-2.0-fast at $0.06, and both cap at 720p. That makes Wan 3.0 the cheapest route to 1080p on the catalog, which is a different claim from cheapest overall and the one that matters if 1080p is your delivery format.

A newer generation that costs half the model it replaces is not the normal pattern. It usually goes the other way, as Gemini 3.8 Flash showed at the same price for 30% more output tokens. Here the direction is reversed and the reason is visible in the spec sheet rather than the marketing: Wan 3.0 collapses what used to be several models into one. Wan 2.7 added first-frame, first-and-last-frame and video continuation as separate task types on top of 2.6. Wan 3.0 takes text, images, video and audio in a single request and covers text-to-video, image-to-video, reference-to-video, editing and continuation without switching model strings.

The full Wan 3.0 vs Wan 2.7 comparison covers what else moved between the two, including the duration ceiling doubling from 15 to 30 seconds, and what to re-check when you swap the model string. Against ByteDance rather than the previous generation, Wan 3.0 vs Seedance 2.5 is the relevant comparison: Seedance costs 2.2x per second but accepts two and a half times as much reference material.

What Wan 3.0 Actually Takes as Input

Up to 10 reference images, 5 reference videos and 5 reference audio clips in one request, with the video and audio each capped at 15 seconds total. Those caps are from Alibaba’s Wan 3.0 API reference, which is also where the duration rules live: an integer in the range 2 to 30 seconds, or -1 for intelligent duration, and when you pass input video the input plus output duration together cannot exceed 30 seconds.

Two defaults are worth knowing before your first call:

• Audio is on. The reference sets audio output to true by default, so the returned file has sound. Ofox exposes generate_audio if you want silence.
• Aspect ratio is adaptive. The supported set is 16:9, 4:3, 1:1, 3:4 and 9:16, with adaptive as the default, so the model picks unless you pin it.

Ofox’s model object reports the same envelope: min_duration_seconds: 2, max_duration_seconds: 30, supports_audio: true, resolutions 480p/720p/1080p with default_resolution: 1080p, and modes t2v, i2v, v2v.

Wan 3.0 Prime: Paying 1.6x for Speed

Prime is the same capability, faster, at $0.08 per second instead of $0.05. Alibaba describes wan3.0-video-prime as the high-speed version, “能力对齐标准版，端到端速度显著提升”, which reads as capability aligned to the standard version with significantly improved end-to-end speed. Note what that sentence does and does not claim: it claims speed, and it claims parity, not superiority, on capability.

So Prime is a latency purchase. At 1.6x the rate, it makes sense for interactive products where a user waits on the clip, and makes no sense for batch pipelines where nobody is watching the queue. There is no published per-model latency figure to compare against, so the honest recommendation is to run both on your own prompt and time them before committing a pipeline to the higher rate.

How to Call Wan 3.0

Video generation is asynchronous everywhere: create a task, then poll. Alibaba’s own flow is documented as “创建任务 -> 轮询获取”, create the task then poll for the result, with the task_id and the resulting video URL each valid for 24 hours and a default query rate of 20 RPS.

Through Ofox the model sits on /v1/videos:

curl -X POST https://api.ofox.io/v1/videos \
  -H "Authorization: Bearer $OFOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "alibaba/wan-3.0",
    "prompt": "A paper boat drifting down a rain gutter, close on the water line",
    "duration": 5,
    "resolution": "1080p",
    "aspect_ratio": "16:9"
  }'

That returns a task you poll for status. The parameters Ofox lists on this model are prompt, duration, resolution, aspect_ratio, size, generate_audio, seed, frame_images, input_references, callback_url and provider. Passing frame_images selects first-frame or first-and-last-frame mode, input_references selects reference-to-video, so the mode follows from what you send rather than from a separate model string. callback_url is the alternative to polling if you would rather be told than ask; the video API polling guide covers the wait-time behaviour either way.

If a request comes back rejected instead, the Wan 3.0 error reference has the literal response body for each failure, including the duration range, the allowed aspect ratios and the model-string mismatch between Alibaba’s wan3.0-video and the gateway’s alibaba/wan-3.0.

Calling Alibaba directly, the endpoints are regional:

https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis
https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis

Both need a Model Studio API key and a workspace ID in the hostname.

Which Route to Pick

• Shipping 1080p: the flat $0.05 rate is roughly 70% under Alibaba’s list price and roughly 58% under the discounted rate, and it does not expire on 23 September.
• Shipping 480P at volume: Alibaba’s discounted 0.21 yuan per second is cheaper than $0.05 while the discount lasts. Re-check on 24 September, when the rate returns to 0.3.
• Mixed resolutions in one pipeline: the flat rate removes a variable. You can raise output quality without re-forecasting the bill.
• You need the Beijing region or a Chinese billing entity: call Alibaba directly. The regional endpoints and workspace model are theirs, not something a gateway reproduces.

The Date to Watch

Alibaba’s 30% discount runs to 23 September 2026. After that the 1080P rate returns to 1.2 yuan per second, which widens the gap against the flat $0.05 rather than narrowing it. Any cost model built on the discounted rate needs a second row for 24 September onward. The Ofox rate carries no equivalent expiry today, though the live catalog at GET https://api.ofox.io/v1/models remains the authority on what you are actually billed.

Sources

• https://help.aliyun.com/zh/model-studio/wan3-video-generation-api-reference
• https://www.ithome.com/0/993/396.htm
• https://ofox.io/models/alibaba/wan-3.0

Alibaba pricing and the discount window are from the 24 August 2026 launch coverage; API behaviour is from Alibaba’s Wan 3.0 API reference. Ofox rates and model attributes were read from the live /v1/models endpoint on 4 September 2026. Yuan-to-dollar comparisons are approximate and move with the exchange rate; the Ofox-to-Ofox multiples do not.