Reading MiniMax H3's per-resolution price table

Updated 2026-10-02

MiniMax H3 is not one price per second. Each host lists its own set of resolution tiers, and each tier has its own price. That is why a single "cheapest host" answer is incomplete, and why the best host at one tier is often not the best at another. This page explains the mechanics and how to read the table on each model page.

The basic model: seconds times a tier rate

Cost for a job is the requested duration times the per-second rate of the tier you asked for, plus the platform's 2% fee. It is charged once at creation, and polling is free. A longer clip scales cost linearly, and a higher tier changes the rate. So you have two independent dials, and the tier is the one people forget to check.

The live comparison

The table below is rebuilt from the live catalog. It shows, for each model on this site, the tier compared, the cheapest and the priciest host, and the gap:

Per second, before VideoRouter's 2% platform fee. For tiered models each row compares the resolution tier with the widest host-to-host gap. Built 2026-10-02 from the live catalog.

Each row compares one tier. The model pages go further and show every host against every tier.

How to read a model page

Open any H3 page in the models section. You will see a grid:

• Rows are hosts. Several providers sell the same checkpoint.
• Columns are tiers such as 480p, 768p, 1080p, 1440p and 2160p, plus labels like 1080p-esr. The -esr and -sr variants are enhanced-super-resolution versions that a host lists as separate priced tiers, so they are not interchangeable with the plain tier of the same number.
• A dash means the host does not offer that tier. Coverage is uneven: some hosts list many tiers and others only a couple.
• A "% off" note compares a price with a stated reference host on that page. Read the footnote to see which.
• The markup line. The page shows prices with no markup beyond the 2% platform fee, so the numbers are the host's own.

Why the spread differs by tier

You cannot see a host's cost structure, but you can observe the patterns that explain the table:

1. Coverage. A tier offered by only two hosts has a narrow range by construction. A tier offered by a dozen has more room for a low and a high outlier.
2. Hosts price tiers independently. One host may discount a low tier heavily to attract experimentation while pricing a high tier close to the list rate. A discount that exists at one tier says nothing about another.
3. Rates are not a fixed multiple. Going up a tier does not scale every host's price by the same factor, so the ranking of hosts can reorder between tiers.
4. Variants of the same label. A tier and its super-resolution variant can be priced differently even at one host.

The practical lesson is to read the column for the tier you intend to use, not the headline number.

Choosing a tier

A reasonable workflow: draft at a low tier, review, then re-run only approved shots at the delivery tier. If you also vary the model, see the image-to-video guide for how input choices interact with resolution.

Check that you got the tier you paid for

Per the video docs, a resolution a model does not support is ignored and the model's default is used instead, without an error. So after any change to hosts or tiers:

import subprocess, json

def probe(path):
    out = subprocess.run(
        ["ffprobe", "-v", "error", "-select_streams", "v:0",
         "-show_entries", "stream=width,height", "-of", "json", path],
        capture_output=True, text=True, check=True).stdout
    s = json.loads(out)["streams"][0]
    return s["width"], s["height"]

Compare the dimensions with what you requested, and check the host's column on the model page for that tier before pinning. Pinning to a host that lacks the tier is the most common way to end up with an unexpected resolution.

Routing and tiers together

Unpinned requests go to the cheapest healthy host. When you request a tier, ask whether the routing and the tier agree: if your target tier is offered by only a few hosts, failover has fewer places to go. If you need a tier consistently, use "provider": {"only": ["host-a", "host-b"]} with the hosts whose column covers it, and keep fallbacks on so you still fail over between them. The Python tutorial shows the exact request shapes.

To estimate spend, multiply your clip count by seconds by the rate in your tier's column, then add a retry allowance for rejected attempts. Re-check the live table when you do, since rates move. Get a key at videorouter.sh/signup, or start from the quickstart.

Frequently asked questions

Keep reading

• MiniMax H3 Providers Compared — How to Pick a Host
• MiniMax H3 Image-to-Video API: start_image_url Guide
• MiniMax H3 Python Tutorial: Create, Poll, Download, Pin
• MiniMax H3 Prompt Guide: A Testable Structure for Video Prompts