Hailuo 3 API and Pricing, Explained Through a Real Pipeline

Hailuo 3 is MiniMax's open-weights video model, usually written as H3, and it is one of the few frontier video models whose price you can work out with a multiplication. ByteDance meters Seedance in opaque video tokens. Hailuo 3 publishes a separate per-second rate for each of the four resolutions it renders, so the cost of a clip is the rate times the seconds, and nothing else.

This post covers what each configuration costs, the one setting that decides most of your bill, what the endpoint refuses to take, and how to call Hailuo 3 as a published API endpoint. Every parameter and price below is read off Treza's live video model catalog as of September 2026.

What you get before you talk about price

A 4K ceiling is rare, and rarer still on a model that also runs to 15 seconds. What H3 is good at is control: it is built for instruction-guided generation, where you describe the change rather than re-roll the idea, and it holds on-screen text and brand marks legible in the frame. That last part is unusual enough among video models to be the reason to pick it. Full capability detail lives on the Hailuo 3 model page.

How Hailuo 3 bills

Per second of output, at a rate set by the resolution you asked for:

Those are provider rates you can multiply. A 10 second clip at 768p is 60 cents of provider cost, and the same 10 seconds at 4K is $1.60. Duration is linear and boring, which is the point. The lever that matters is the resolution dropdown, because the gap between 768p and 4K is a factor of roughly 2.7 on the same shot.

The resolution dropdown is the whole pricing decision

The provider defaults this endpoint to 2K. We default the node to 768p instead, and that choice is most of the difference between a cheap model and a mid-priced one. A 9:16 Short on a handset does not use 2K, and neither does a 720p broadcast. Defaulting to 2K means paying 13 cents per second for pixels the encode throws away, on every draft, forever.

We learned that the expensive way. Hailuo 3 used to be reachable through a second lane, and measured on 2026-08-29 that route took 4 minutes 23 seconds and $0.78 for a single 6 second clip, because the API exposed no resolution parameter and quietly defaulted to 2K. We hid the duplicate rather than leave a slower, pricier path to the same picture. The lesson generalizes: on any video endpoint, find out what resolution it defaults to before you decide what it costs.

So draft at 768p and raise it only on the takes shown large, on a site hero, a paid placement, or a desktop display. The 480p tier saves one cent per second, which is rarely worth what it costs in the picture.

What you actually pay on Treza

Hailuo 3 through a Treza pipeline is metered from a prepaid credit balance. One credit equals $0.01, credit packs start at $5, credits never expire, and there is no subscription. A failed generation charges nothing.

Before a run starts, the platform estimates the media cost of the graph and refuses to start one your balance cannot cover. For Hailuo 3 the estimator reserves 16 cents per second, the model's 4K rate, because the hold has to cover the most expensive configuration the node could be set to. Drafting at 768p, that is roughly two and a half times what the run will cost. Afterward the run record carries the real cost and the difference returns to the balance. Reserves are holds, not prices.

For sizing a budget across the catalog, the figure on the pricing page is $1.06, the median charge for one video generation across all models. Hailuo 3 drafts at 768p land well under it and a 15 second 4K render lands above it. The cost breakdown post walks a finished video end to end, with narration, transcription, and captions in the graph rather than just the render.

H3 Max, the draft node

MiniMax H3 Max is the post-trained H3, and it sits in the same picker. It trades resolution for throughput: 480p and 768p only, 5 to 10 seconds, at 5 and 8 cents per second, with the provider claiming a 5 second 768p clip in under 3 seconds.

At 768p, Max costs 8 cents per second against H3's 6, so you pay a third more and what you buy is wall clock. That is the right trade when you want to see ten framings before committing, and the wrong one when you already know the shot. Iterate on Max, render the keeper on H3, and switching is one dropdown.

What the endpoint will not take

Hailuo 3 is widely described as having keyframe control. That describes the model. The endpoint decides what you can send, and on the text to video endpoint we run you send a prompt, a duration, an aspect ratio, and a resolution. No first or last frame keyframes, and no reference images.

That matters for two jobs. If a clip has to open or close on an exact frame, use Seedance 2.5, which takes an image at each end and travels between them. If a character or product has to look the same across separate generations, Seedance 2.0 and Seedance 1.5 Pro accept the multi-image reference input.

Native audio is not a switch either. It is part of the model rather than a parameter, so it arrives with every clip and there is no silent-draft discount to be had.

The full parameter-level comparison against our default model is in the Hailuo 3 versus Seedance 2.5 post.

Calling Hailuo 3 as an API without building the integration

Direct API access to a frontier video model is never one HTTP call. It is credentials, an asynchronous submit-and-poll job pattern, retries for transient failures, moderation rejections, and somewhere to put the output before your product can use it.

On a Treza AI video pipeline, Hailuo 3 is a node on a canvas. Set the duration, aspect ratio, and resolution, wire it next to a script step, a voiceover step, or a captioning step, and publish the whole chain as one versioned endpoint. Your application calls a typed /invoke route with JSON in and JSON out, or points an existing OpenAI SDK at the OpenAI-compatible route.

The swap is what pays off later. When the next model wins the leaderboard, you change the dropdown and republish. The endpoint your product calls does not change and nobody ships a migration. The only thing to check is whether the new model supports the durations, aspect ratios, and resolutions your node was set to. Our hosted MCP server exposes the same pipelines as tools, so rendering a brief on Hailuo 3 and returning the link is something an agent carries out rather than describes.

Where Hailuo 3 earns its place in a graph

Three jobs, from our own pipelines. Product and brand beats, where legible labels and on-screen words matter more than clip length, which is the product video case. Anything shown large, because 2K and 4K at 13 and 16 cents per second is the cheapest sharp frame in the catalog. And short vertical cuts, where 6 cents per second with audio included is hard to beat as long as the segment fits inside 15 seconds and needs no pinned frames.

The fastest way to answer what it costs for your own work is to run your real prompt at your real duration and read the charge. Open the AI video generator, pick Hailuo 3, set the resolution the placement needs, and generate once.

Frequently Asked Questions

How much does Hailuo 3 cost per second?

Five cents at 480p, six cents at 768p, thirteen cents at 2K, and sixteen cents at 4K, as published for the endpoint we run as of September 2026. Multiply by the seconds for the provider cost of a clip. On Treza that is charged from a prepaid credit balance, and a failed generation charges nothing.

What is the longest clip Hailuo 3 can generate?

Fifteen seconds in a single generation, at any whole number of seconds from 5 to 15. For longer pieces, stitch generations with the Sequence node, which handles up to 12 shots, or assemble them in the timeline editor.

Does Hailuo 3 output 4K?

Yes, at sixteen cents per second. The node defaults to 768p rather than the provider's own 2K default, because most output here is phone-first and the extra pixels get discarded in a 720p encode. Resolution is a dropdown, so 2K and 4K are one click away.

Does Hailuo 3 support first and last frame keyframes?

Not on the endpoint we run. It takes a prompt, a duration, an aspect ratio, and a resolution, with no keyframes and no reference images. If a clip has to start or land on an exact frame, use Seedance 2.5, which supports both ends.

What is the difference between Hailuo 3 and H3 Max?

H3 Max is a post-trained variant of the same weights, tuned for speed: 480p and 768p only, 5 to 10 seconds, at five and eight cents per second, with the provider claiming a 5 second 768p clip in under 3 seconds. Iterate on Max, render the keeper on H3.

Why does my credit balance reserve more than the clip costs?

The estimator reserves sixteen cents per second for this model, its 4K rate, so a run can never start that the balance cannot finish. The run settles at actual provider cost plus markup and the difference returns to the balance. A draft at 768p settles well under its reserve.

Can I call Hailuo 3 through an API without a MiniMax account?

Yes. On a Treza pipeline, Hailuo 3 is a node you configure once and publish as a versioned HTTP endpoint, with a typed /invoke route and an OpenAI-compatible route. A hosted MCP server exposes the same pipelines as tools for agents. There is no subscription: generations are metered from prepaid credits that start at $5 and never expire.

Is Hailuo 3 open weights?

Yes, MiniMax released H3 as an open-weights model. On Treza you use it through the same picker and per-second pricing as every other model, with no hosting to manage.