One prompt, two models: asyncio.gather Wan and Seedance for $21.08

Submitting one 30 s prompt to both wan-3.0 and seedance-2.5 at 720p reserves $3.75 + $17.334 = $21.084, and asyncio.gather sends both requests at once. Give each request its own Idempotency-Key: reusing one key with a different body returns 409 idempotency_conflict.

What the comparison costs

Both models accept 30 s at 720p. Seedance is token-priced, so its total depends on resolution, length and aspect, and the figure below is for 9:16. Wan 3.0 is a flat per-second rate.

Run it once per prompt to choose a model, not on every request; the Seedance side is 4.6 times the Wan side.

The script

Python has no top-level await in a script, so main() runs under asyncio.run. The result lists the job id and polling_url for each model; poll them as in the quick start.

import asyncio, os, uuid
import httpx

H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
PROMPT = "A street food cart at dusk, steam rising, handheld camera"
BODIES = [
    {"model": "wan-3.0", "prompt": PROMPT, "duration": 30, "resolution": "720p"},
    {"model": "seedance-2.5", "prompt": PROMPT, "duration": 30,
     "resolution": "720p", "aspect_ratio": "9:16"},
]

async def submit(client, body):
    r = await client.post(
        "https://api.sume.com/v1/videos", json=body,
        headers={**H, "Idempotency-Key": str(uuid.uuid4())},
    )
    return body["model"], r.status_code, r.json()

async def main():
    async with httpx.AsyncClient(timeout=30) as client:
        for model, code, data in await asyncio.gather(*(submit(client, b) for b in BODIES)):
            print(model, code, data.get("id"), data.get("polling_url"))

asyncio.run(main())

Gotchas

Two paid jobs run at once, so on the Free plan (1 processing, 5 queued) a second pair of submits can wait in the queue, and a seventh paid job gets 429 queue_full. If one side returns 402, the other may already be reserved; check each status code, not just the first.

Keep the generated key with the model name in your own log, so a retry after a timeout reuses it and does not pay twice.