Is Kandinsky 6.0 Video on Sume? No, and the closest audio ids

No. Kandinsky 6.0 Video is not in Sume's video catalog, and Kandinsky Lab's own repository lists no hosted API for it (both read 2026-10-11). If you want a 5-second clip with synchronized sound and an optional first frame through one API call, Sume's closest ids are seedance-2-mini, seedance-2-fast, seedance-2, seedance-2.5, kling-3, wan-3.0, minimax-h3 and gemini-omni-flash-1.1.

That answer is narrow on purpose. Kandinsky 6.0 shipped as downloadable weights on 2026-10-06, so the real choice is run it yourself or call a hosted model. Sume only covers the second half, and this page shows where the two overlap.

What Kandinsky Lab says it shipped

The numbers below come from Kandinsky Lab and the kandinsky-6 repository, both read on 2026-10-11. Nothing here is a Sume measurement.

What Sume lists instead

Sume's video generation docs say the supported models, capabilities and prices come from GET /v1/videos/models. The ids below come from the catalog code on the main branch, which this page read on 2026-10-11. A row can appear only when its provider is configured, so treat the endpoint as the final word for your workspace.

Where the match breaks

Three differences matter before you swap one for the other.

• Sume does not fix the length at 5 seconds. Every id above takes a duration, so a 5-second clip is a request value, not a model property.
• Sume's grok-imagine-video-1.5 is not on the list. Its catalog row reports no audio and takes one first frame, so it does not match an audio-video model.
• Weights are not available. A hosted id gives you a job, a price and a result URL, but you cannot fine-tune it, pin a checkpoint, or run it offline.
• Seed control is also different. Sume's video docs state that no v1 model accepts a seed, so a hosted clip is not reproducible on request.

Check the catalog from code

This script lists every video id your key can see that reports audio and takes a first frame. It reads data, generate_audio and supported_frame_images as documented on the video generation page.

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"]:
    durations = m["supported_durations"]
    if m["generate_audio"] and "first_frame" in m["supported_frame_images"]:
        print(m["id"], min(durations), max(durations), m["supported_resolutions"])

How to decide

Choose Kandinsky when you need open weights, an MIT license you can audit, and your own GPUs. Choose a Sume id when you want a job queue, a per-clip price and no hardware. If you only want to compare looks, run the same prompt through Kandinsky locally and through a pinned Sume id, and judge the two clips side by side. Pin the id: sume/auto never names the family it ran, so it cannot serve as a comparison arm.