Wan 3.0 vs Seedance 2.5: $0.05 vs $0.11 per Second

Wan 3.0 costs $0.05 per second of output video on Ofox. Seedance 2.5 costs $0.11, which is 2.2x as much. That is the headline, and for most cost-driven pipelines it decides the question. The specs complicate it in exactly two places: Seedance takes two and a half times as much reference material, and Wan cannot do 21:9. Everything below is read from the live Ofox catalog on 4 September 2026.

TL;DR

• Per second, Wan 3.0 is less than half the price. $0.05 against $0.11. A 5-second clip is $0.25 against $0.55, a 30-second clip $1.50 against $3.30.
• Wan 3.0 makes shorter clips. 2-second minimum against Seedance’s 4. Both stop at 30.
• Seedance 2.5 takes far more references. 50 assets (30 images, 10 videos, 10 audio) against Wan’s 20 (10 images, 5 videos, 5 audio).
• Different defaults. Wan defaults to 1080p, Seedance to 720p. The same request with no resolution set costs the same on each route but does not produce the same file.

Wan 3.0 vs Seedance 2.5: The Spec Table

Prices are the output_video_per_second field on the live /v1/models response; the duration, resolution, aspect and audio rows are the video_attributes object on each model. Read 4 September 2026.

One caveat on the resolution row. Seedance 2.5’s catalog description text says 480p/720p, but its video_attributes.resolutions array lists 1080p as well. The structured field is the one the API validates against, so 1080p is available; the prose is stale. Worth knowing before you assume a capability is missing because a description did not mention it.

The Price Gap Is the Whole Argument

At 2.2x, the cost difference is larger than any spec difference on this table. Video generation bills by the second, so the multiple applies to every clip you ever generate, unlike a token-priced model where output length varies by prompt.

Nothing in that table is a surprise once you know the per-second rates, which is the point: the arithmetic is linear and there is no caching, no batch discount and no thinking-token variance to model. Whatever you pick, you pay the rate times the seconds.

That makes the decision unusually clean. If the two models produce output you consider interchangeable for your use case, take the cheaper one. The rest of this post is about the cases where they are not interchangeable.

Where Seedance 2.5 Earns the Premium

A reference library of 50 assets against Wan’s 20, and 21:9. Those are the two capabilities Wan 3.0 does not match.

Seedance 2.5 accepts up to 30 reference images, 10 reference videos and 10 reference audio clips in a single request. Wan 3.0 accepts 10, 5 and 5, and additionally caps reference video and reference audio at 15 seconds each in total. If your product feeds a character sheet, a location set and a motion reference into every generation, that ceiling is a hard constraint rather than a preference, and it is reached quickly.

The aspect ratio list is the smaller difference but a real one: Seedance lists 21:9, Wan does not. For cinematic letterbox output that is decisive, and cropping a 16:9 render to 21:9 costs you vertical resolution rather than being free.

Where Wan 3.0 Wins Beyond Price

Clips under 4 seconds, and 1080p as the default. Wan 3.0’s minimum duration is 2 seconds; Seedance’s is 4. If you generate short loops, stingers or reaction clips, Seedance’s floor means generating 4 seconds and trimming, paying for footage you discard. At $0.11 per second, discarding 2 seconds per clip costs $0.22 every time.

The default resolution difference is a smaller trap. Send an identical request with no resolution set to each model and Wan returns 1080p while Seedance returns 720p. That is not a billing difference on this route, since both bill per second regardless of resolution, but it is a quality difference that will not announce itself in an A/B comparison unless you pin the resolution on both.

The Calls Are Nearly Identical

Both models sit on the same Ofox endpoint and take the same parameter names, which makes an A/B cheap to run:

import os, requests

OFOX = "https://api.ofox.ai/v1/videos"
HEAD = {"Authorization": f"Bearer {os.environ['OFOX_API_KEY']}"}

prompt = "A paper boat drifting down a rain gutter, close on the water line"

for model in ["alibaba/wan-3.0", "bytedance/seedance-2.5"]:
    r = requests.post(OFOX, headers=HEAD, json={
        "model": model,
        "prompt": prompt,
        "duration": 5,
        "resolution": "1080p",     # pin it; the two models default differently
        "aspect_ratio": "16:9",
    })
    print(model, r.status_code, r.json().get("id"))

Both expose 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 on either; input_references selects reference-to-video. Both are asynchronous: you get a task back and poll it, or supply callback_url and be told instead. The polling and wait-time guide covers what to expect while the task runs.

Pin resolution explicitly in any comparison. Leaving it unset compares 1080p Wan against 720p Seedance and tells you nothing.

Pick Wan 3.0 When

• Cost per clip is the constraint. At 2.2x, nothing else on the spec sheet outweighs it at volume.
• You need clips shorter than 4 seconds.
• 1080p is your normal output and you would rather it be the default than a parameter you must remember.
• You want first-frame, first-and-last-frame, reference and continuation modes without switching model strings. The Wan 3.0 pricing and access guide covers that unified-model behaviour and the Alibaba-direct route.

Pick Seedance 2.5 When

• You feed more than 10 reference images, or more than 5 reference videos or audio clips, per generation.
• You need 21:9 output.
• Your prompts are already tuned against ByteDance motion behaviour and re-tuning would cost more than the price gap saves. The Seedance 2.5 prompt guide is the reference for that tuning.

When Neither Is the Answer

If you are still on Wan 2.7 or 2.6, both bill $0.10 per second on Ofox, so Wan 3.0 halves your rate rather than raising it. That is an unusually easy upgrade and the Wan 3.0 pricing guide has the migration detail. If your requirement is 4K, neither of these two lists it, and the older Seedance 2.0 flagship tier is where that capability sat.

Sources

• https://ofox.ai/models/alibaba/wan-3.0
• https://ofox.ai/models/bytedance/seedance-2.5
• https://help.aliyun.com/zh/model-studio/wan3-video-generation-api-reference

Prices, duration bounds, resolutions, aspect ratios and reference-asset ceilings were read from the live Ofox /v1/models endpoint on 4 September 2026. Per-second rates on other routes, including calling Alibaba or ByteDance directly, differ and are covered in the linked pricing guide.