A new AI video model launched: check Sume's catalog in one call

To see whether a newly announced video model is usable on Sume, call GET /v1/videos/models and look for its id; the response lists each model's supported_resolutions, supported_durations, supported_frame_images, supported_input_references and generate_audio. If the id is not in data, it is not callable. As read on 2026-10-02, seedance-2.5, wan-3.0 and minimax-h3 are Video Router ids, and HappyHorse 1.1 and Luma Ray3.2 are not.

The shape comes from Sume's Video generation docs and the Video Router docs, read 2026-10-02.

What does the one-call check look like?

The script below prints each model's id, resolutions and shortest and longest duration. It needs only an API key in SUME_API_KEY.

Run it the day a model is announced, then again a week later; a catalog changes when Sume adds an id, not when a vendor posts.

import os
import requests

r = requests.get(
    "https://api.sume.com/v1/videos/models",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    timeout=30,
)
r.raise_for_status()
for m in r.json()["data"]:
    d = m.get("supported_durations") or []
    span = (min(d), max(d)) if d else None
    print(m["id"], m.get("supported_resolutions"), span)

Which fields should I read before I plan around a model?

Read the fields that can break a plan, not the description string. Docs state that limits are not uniform across models, so one envelope cannot be assumed.

The Video Router catalog (GET /v1/video-router/models) adds capabilities and the billing fields, including a billable_margin per row.

What if the id is missing but the vendor says an API exists?

The vendor's API and Sume's catalog are separate things. Luma states that an API is available for Ray3.2, and Hedra lists HappyHorse 1.1 on Alibaba Cloud Model Studio and partner APIs, both read 2026-10-02. Neither makes the id callable on Sume.

A request with an unknown model id fails; Sume's admission docs list 404 model_not_found for a model that does not exist in your workspace. Use the closest catalog id instead, as in which of these models run on Sume.

Put the script on a schedule if you track launches: diff today's ids against yesterday's, and open a ticket for any new id with a duration or resolution that fits a job you already have. That turns a news cycle into a short list of things to test.