New AI video models in 2026: which ones have a Sume model id

Of the models a video tracker lists for 2026, Sume's Video Router catalog has ids for Seedance 2.5, Seedance 2.0, MiniMax H3, Kling 3 and Wan 3.0. It has no id for LTX-2.5 or Veo 3.1. The tracker also records that the OpenAI Sora API ended on 2026-09-24 (Magic Hour tracker, read 2026-10-06), so this list is the one people reach for when they pick a new target.

The catalog moves, so treat the table as a dated reading and confirm with GET /v1/videos/models before you pin anything. Competitor facts below come from the tracker page; Sume facts come from the Sume docs.

The table

How to read it

• A date on the tracker is the vendor's release date, not the date it reached Sume. The Sume column says what the docs list today, nothing more.
• Tracker length and Sume length can differ. The tracker says MiniMax H3 is 15 s at 2K; Sume's doc says native 480p and 768p, with 2K and 4K upscales billed if you request them.
• Seedance 2.5 goes to 30 s on Sume, which is the longest single request in the catalog together with Wan 3.0.
• For LTX-2.5 and Veo 3.1, the honest answer is that Sume does not list them. The existing posts on those two ask the same question in more detail.

Check the catalog yourself

This prints every id Sume lists with its duration range, so you can compare it with any tracker without trusting a blog table.

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 sorted(r.json()["data"], key=lambda x: x["id"]):
    d = m.get("supported_durations") or []
    span = f"{min(d)}-{max(d)} s" if d else "n/a"
    print(f"{m['id']:28} {span:10} audio={m.get('generate_audio')}")

If the model you want is not in the output, do not force a similar id into your code. Pick a listed model, run one prompt through it, and compare. The early October re-test list says which listed models to try first, and the LTX and Kling catalog check shows the same method on two names.

What to do about a model that is not listed

Two entries in the table have no Sume id: LTX-2.5 and Veo 3.1. If your product depends on one of them, say so plainly to your users and your team: it is not on Sume today, and a similar id is not the same model. A silent swap changes the look of your output and the numbers on your dashboard at the same time, which makes the cause hard to find later.

If you can accept another model, pick the listed one whose limits match your jobs, not the one with the most familiar name. The useful limits are duration range, resolutions, reference types and aspect ratios, all of which the catalog returns per model. The tracker page is a good source for what exists in the market; the Sume catalog is the only source for what you can call.

• Write the list of models you need, with the limits that matter for each, before you read any tracker.
• Match that list against GET /v1/videos/models, not against a blog table that may be stale.
• For each gap, either choose a listed model and test it on twenty of your own prompts, or keep that workload where it is.
• Date every comparison you keep. A catalog is a moving target, and a result without a date is hard to trust a month later.