Check one video model first: GET /v1/video-router/models/{id}

The cheapest failed video request is the one you never send. Sume's Video Router has a one-model endpoint for exactly this: GET /v1/video-router/models/{model_id}. It returns the same row as the catalog list, and answers 404 when the id is not in the catalog.

What the row tells you

The Video Router docs say the limits differ by model, so read capabilities and do not assume one envelope.

• Resolutions, durations and which of first_frame, last_frame, image, video and audio references the model accepts.
• Whether audio is a toggle or always on. MiniMax H3 and Gemini Omni Flash 1.1 have native audio with no toggle.
• Price: the row's per-model rate already includes Sume's 1.25 multiple, and billable_formula is the per-model truth.

Examples that matter

A guard in Python

This helper fetches one model's row and raises when the id is not in the catalog. Print the row, then compare your length against the duration range it reports before you submit. The response envelope uses data; adapt to the row you receive.

import os

import requests

BASE = "https://api.sume.com/v1/video-router"
HEADERS = {"Authorization": "Bearer " + os.environ["SUME_API_KEY"]}


def model_row(model_id):
    r = requests.get(BASE + "/models/" + model_id, headers=HEADERS, timeout=30)
    if r.status_code == 404:
        raise ValueError(model_id + " is not in the Sume catalog")
    r.raise_for_status()
    return r.json()


row = model_row("seedance-2.5")
print(row)

Try it with an id from a launch announcement. If it 404s, the model is not available through Sume yet, whatever its vendor page says.