MiniMax says H3 is 2K: what Sume lists, 768p to 1080p
MiniMax's own page puts H3 at 2K and 15 s. Sume lists native 768p for H3 and H3 Max, 1080p refinement on Max, and priced 2K/4K upscale SKUs on H3. Prices.
MiniMax describes H3 as a model that generates video with native stereo sound at up to 2K and up to 15 seconds. On Sume the two catalog ids, minimax-h3 and minimax-h3-max, are built around native 768p, with 1080p as a refinement on Max and 2K/4K as priced upscale rows on minimax-h3.
So if you search for a 2K Hailuo-class API, the answer is: you can order 768p natively on both ids, 1080p on H3 Max, and read the catalog before you assume 2K. This post lines up what the vendor says with what Sume documents.
What MiniMax says on its own page
MiniMax's H3 announcement says H3 "can generate video with native stereo sound" and gives the ceiling as "up to 15 seconds at 2K resolution" (MiniMax H3 announcement, read 2026-10-05). The same page says its 2K path has the H3 base model regenerate its own low-resolution output in context, rather than run a classic super-resolution filter.
MiniMax's open-source news page (read 2026-10-05) is more specific about what ships to the public: the open H3-Base release uses 768 pixels on the shorter side by default, and the 2K module is described as not open-sourced. In other words, the vendor itself separates a 768p base pass from a hosted 2K stage.
What Sume documents for the two ids
Sume's Video Router catalog lists minimax-h3 at 480p and 768p for 5 to 15 seconds, with native stereo audio and no audio toggle. The docs state that 768p is native, not 720p, and that 2K/4K upscales are priced if a request includes them. minimax-h3-max is listed at 480p, 768p and 1080p, where 1080p is a latent refinement from 768p, and the docs say there is no 2K or 4K on that id.
Because limits differ per model, the docs tell callers to read capabilities from GET /v1/video-router/models instead of assuming one envelope. That is the check to run before you hard-code a resolution.
| Source | Resolution claim | Duration | Audio |
|---|---|---|---|
| MiniMax H3 page | Up to 2K, 768p tier also described | Up to 15 s | Native stereo |
Sume minimax-h3 | 480p, 768p native; 2K/4K upscales priced if requested | 5-15 s | Native stereo, always on |
Sume minimax-h3-max | 480p, 768p native; 1080p latent refinement; no 2K/4K | 5-15 s | Native stereo, always on |
What each tier costs on Sume
Sume bills the provider list price times 1.25. The documented provider list prices per second are: H3 at $0.05 (480p) and $0.06 (768p); H3 Max at $0.05, $0.08 and $0.16 for 480p, 768p and 1080p. The pricing source also carries 2K at $0.13 and 4K at $0.16 per second for H3 upscales. With the 1.25 multiple, a 10-second clip bills as below.
| Id and tier | List per second | Sume per second | 10 s total |
|---|---|---|---|
minimax-h3 768p | $0.06 | $0.075 | $0.75 |
minimax-h3-max 768p | $0.08 | $0.10 | $1.00 |
minimax-h3-max 1080p | $0.16 | $0.20 | $2.00 |
minimax-h3 2K upscale | $0.13 | $0.1625 | $1.63 (rounded up to the cent) |
minimax-h3 4K upscale | $0.16 | $0.20 | $2.00 |
How to choose
If the delivery target is a vertical social ad, 768p native is usually the point where you compare motion and prompt adherence, and Max's 1080p refinement doubles the per-second price from $0.10 to $0.20. If the asset has to be 2K or larger, the only documented route on these two ids is the H3 upscale pricing, which Sume bills only when a request includes it. Confirm the live supported_resolutions for the row first, since the catalog is the source of truth.
- Need native stereo audio: both ids, no toggle, no
generate_audioswitch. - Need 1080p:
minimax-h3-max, priced at $0.20 per second. - Need more than 15 seconds: neither id; the docs cap both at 15 s.
A worked example: a 12-second product clip
Say you need a 12-second product clip with sound. On minimax-h3 at native 768p it bills 12 x $0.075 = $0.90. On minimax-h3-max at 768p it bills 12 x $0.10 = $1.20, and at 1080p it bills 12 x $0.20 = $2.40. The 2K upscale on minimax-h3 would be 12 x $0.1625 = $1.95 after rounding up to the cent, but only if the live catalog still lists that resolution for the row.
None of these prices include extra reference-image charges. The docs say minimax-h3 adds $0.08 list per reference image after the fifth, so a brief with seven references adds 2 x $0.10 = $0.20 on that id.
What to verify before you ship
Treat any vendor page as a statement about the vendor's own hosted product and any Sume doc as a statement about what Sume lists. They can both be true at once: MiniMax can offer 2K on its platform while a given Sume row lists 768p and 1080p only. When the two differ, the Sume catalog response is what your request is validated against.
- Call
GET /v1/video-router/models/minimax-h3and readcapabilitiesfor the live resolution list. - Send the explicit
resolutionstring, for example768p; the docs say 720p is not the same tier on H3. - Keep the submit response, where
usage.billable_amount_usd_microscarries the price you were quoted.
Sources
Related posts
More in Models
- Nano Banana 2 takes 14 references; Sume's catalog allows 10
Google lists 14 reference images for Nano Banana 2. Sume's descriptor for it is 10. How to trim, pack or choose another model before the 400.
- Nano Banana 2 Lite takes 14 references; Sume's cap is 10
Google's docs give Nano Banana 2 Lite 14 reference images in total. Sume's input_references cap is 10 (16 on ChatGPT Image 2.5), and Lite isn't in its catalog.
- Nano Banana 2 Lite is 1K only: send 2K and 4K to NB2
Google lists Nano Banana 2 Lite at 1K only; NB2 offers 512px to 4K at $0.045 to $0.151 per image. On Sume, resolution accepts 512, 1K, 2K or 4K per model.
- Nano Banana 2 Lite has no Search grounding; NB2 does
Google's docs say Nano Banana 2 supports Google Search and image search grounding; Lite does not. Sume has no grounding flag and no Lite row.
Written by Sume