What a per-second rate does and does not include, and which product you are actually calling.
Two notes about reading a hosted video model's price list correctly: what stays constant across aspect ratios, and why the thing sold as one model is usually two checkpoints with different appetites.
A per-second rate looks like the simplest pricing model in software, and it is, right up to the point where you try to use it to predict a bill. Two things trip people up, and both are visible in the documentation if you know which line to read.
The first is that aspect ratio is not a price lever. A nine-by-sixteen frame and a twenty-one-by-nine frame cost the same per second, which sounds like trivia until you notice how many production decisions are quietly made on the assumption that vertical is cheaper. It is not, and knowing that changes the order you produce things in.
The second is that the thing you are billed for as one model is frequently two. Most open-weight video systems ship separate checkpoints for different input shapes — one that takes frames, one that takes references — and they do not accept the same inputs. Calling the wrong one is the most common cause of an error message that reads like a bug and is actually a routing mistake.
Neither of these is a secret. Both are in the vendor's own documentation, and the consolidated version of it sits at https://minimax-h3ai.video.
Aspect ratio is not a price lever on a per-second model. What that frees you to do, and the order it changes in a production.
Open-weight video systems ship separate checkpoints for frame inputs and reference inputs. Which one answers your call, and what happens when you pick wrong.