MiniMax H3 or H3 Max on Sume: which id for a 5 to 15 second clip?

Use minimax-h3 when 768p is enough, and minimax-h3-max when you need 1080p. Both ids take 5 to 15 second clips, text-to-video, start and end frames, and the same reference limits; the differences are resolution and price. On Sume, a 10-second clip at 768p is $0.75 on H3 and $1.00 on H3 Max, and 1080p exists only on H3 Max, at $2.00. The ids and limits come from the catalog behind the Video Router docs.

How do the two ids compare?

Prices are the catalog's list rates times 1.25, rounded up per clip.

What do they have in common?

Both ids list the same aspect ratios, adaptive, 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16, which includes 21:9 and is among the widest sets in the catalog. Both always produce native stereo audio with no generate_audio toggle, both reject bitrate_mode, and both take up to 9 reference images, 3 reference videos and 3 reference audio files, with images, videos and audios together at most 12. Audio cannot be the only reference.

What does 1080p actually mean here?

Both models render natively at 480p and 768p; the catalog marks 768p as native and first-class, and says it must not be treated as 720p. On H3 Max, 1080p is a latent refinement from native 768p, with no 2K or 4K, and the catalog adds a note about checking the opening frames of 1080p image-to-video clips for a baked crop jump. If you do not need the extra pixels, stay at 768p and save $1.00 on a 10-second clip.

What does a full 15-second clip cost?

At the top of the range, a 15-second clip on minimax-h3 is $0.94 at 480p and $1.13 at 768p. On minimax-h3-max it is $0.94 at 480p, $1.50 at 768p and $3.00 at 1080p. The 480p rate is the same on both ids, so the cheaper id only wins when you step up to 768p.

Neither id accepts seed, size or provider.options, and neither has a generate_audio switch, so a request that works on one works on the other as long as it stays at 480p or 768p. That makes it easy to draft on minimax-h3 and re-send the same body to minimax-h3-max with only model changed.

Which fields behave the same on both?

Both ids refuse bitrate_mode, and neither exposes sync_mode. Both take reference_image_urls, reference_video_urls and reference_audio_urls, and on both a reference video or audio file must run 2 to 15 seconds with a combined 15 seconds per kind. Because a request that fits one id fits the other, the only decision in the request body is the model and the resolution.

If your brief needs a first and last frame, both ids accept them: the catalog lists end_frame as true for both. If your brief needs a 21:9 cinematic crop, both list it too. So the choice really comes down to the three numbers in the price table and whether 1080p is worth the step.

Which should I pick for which job?

For drafts, vertical social clips and reference-driven tests, minimax-h3 is cheaper than H3 Max at both shared resolutions, and its 768p rate of $0.075 per second is the lowest 768p rate in the catalog for this family. For a clip that will play on a large screen, or that you will crop afterwards, take H3 Max at 1080p. One more catalog line matters if you use many photos: on H3 the first 5 reference images are free and each extra image has an $0.08 list add-on, while H3 Max reference-to-video is billed on output seconds. Read the live catalog entry for the id before you run a 9-image job.