Kandinsky 6.0 5-second audio clip: price it on Sume rows

Kandinsky 6.0 has no price to quote: the project page (read 2026-10-11) lists neither a rate nor a hosted API, and Sume does not carry the model. If you need a 5-second clip with sound today, the closest Sume rows cost $0.1875 (Gemini Omni Flash 1.1 at 360p) up to $0.9375 (the same row at 1080p), and $0.50 for MiniMax H3 Max at its default 768p.

This post prices a 5-second, sound-on clip on the Sume rows that bill audio, so you can set a budget before a Kandinsky-style workload reaches you through some other route.

What Kandinsky 6.0 promises, and what it leaves out

The Kandinsky Lab page (read 2026-10-11) describes two variants, Lite 3B and Pro 29B, under an MIT license. It describes 5-second clips with 44 kHz audio from text-to-audio-video and image-to-audio-video modes, with super-resolution to Full HD at 24 fps. The paper on arXiv (read 2026-10-11) says clips are generated at SD and then upscaled.

What neither page gives is a per-second or per-clip price. With open weights, the cost depends on whoever runs them, so any dollar figure for Kandinsky would be invented. The fair comparison is therefore not a price match but a budget bracket: what does a sound-on 5-second clip cost on rows you can call today?

Sume rows that bill a 5-second clip with sound

The Sume rows below come from the price tables and golden amounts in the Sume repository, which apply the house margin of 1.25 to the list price. Gemini Omni Flash 1.1 always generates audio and takes 3 to 10 seconds. Wan 3.0 takes 2 to 30 seconds and its price does not change with audio. MiniMax H3 Max always has audio and takes 5 to 15 seconds, so 5 seconds is its floor.

Check the live rate in the catalog before you commit a budget; the table is a snapshot.

Which resolution matches Kandinsky's output

Kandinsky generates at SD and upscales to Full HD, so its native detail sits closer to the 360p and 480p rows than to the 1080p rows. If you are judging quality per dollar, compare the upscaled result with the 720p rows rather than with 4K.

A 5-second Omni clip at 720p ($0.625) and a Wan clip at 720p ($0.625) cost the same. Wan at 480p costs $0.3125, the next step up from Omni's 360p tier at $0.1875. Neither row is a like-for-like match for Kandinsky, so test one clip on each before you scale a batch.

Budget for a batch of 40 clips

A concept test often needs dozens of takes. Forty 5-second clips cost 40 times the per-clip figure: $7.50 on Omni at 360p, $25.00 on Omni or Wan at 720p, and $20.00 on H3 Max at 768p. Those totals assume every clip succeeds. Sume's errors and credits page documents the 402 insufficient_credits response you receive if the wallet cannot cover the hold when you submit.

For the exact seconds each row accepts, see the shortest clip each row accepts. For where the other new launches stand against Sume ids, see the video models roundup.

What to do if you want Kandinsky itself

Because the weights are MIT-licensed, you can run Kandinsky on your own hardware or a hosted GPU and bring the finished MP4 into Sume afterwards. Sume's Timeline render accepts a Sume-hosted clip as a source, so the rest of your edit can still be priced at $0.10 per output minute. Treat the Kandinsky step as outside the Sume bill, and size the Sume side using the table above.