H3 Max vs MiniMax H3: Same Price, 35x Faster, No 2K

Most of the coverage of H3 Max since it launched on 1 September has landed on "faster and cheaper." The "cheaper" half is wrong, and it is the half people budget against.

At 768P, H3 Max and MiniMax H3 cost exactly the same per second: $0.08. H3 Max is only cheaper if you step down to 480P, or if you were paying for 2K and no longer need it.

What you actually buy with H3 Max is time. This piece works through the spec sheet, then the two tables that matter: what twenty attempts cost on each, and what twenty attempts take.

One thing to be upfront about: I have not run these two head to head. Everything below comes from published specs, the two API schemas, and arithmetic you can redo yourself. The three-second figure for H3 Max is the published one, and every timing here inherits it.

Specs at a glance

H3 figures come from MiniMax's published API rates and limits; we keep them in what MiniMax H3 is and the H3 API guide. H3 Max figures are from its published model pages and API schemas, checked 5 September 2026. Introductory discounts were running on both the clip endpoints and Director at the time of writing, on different end dates — check the live pricing page before you commit to a number.

Two rows do the work here. Resolution, where H3 goes to 2K and H3 Max stops at 768P. And the cost row that isn't a difference at all.

The price row is the story

This matters because "cheaper and faster" sets up the wrong decision. If you assume H3 Max halves your bill, you will move your whole pipeline across, discover your deliverable needed 2K, and end up paying twice — once to draft, once to re-render.

Where the money genuinely changes:

• Dropping from 2K to 768P takes you from $0.13 to $0.08 a second. That is a 38% cut, and it is a resolution decision, not a model decision. H3 at 768P gets you the same saving.
• Dropping to 480P takes you to $0.05, which H3 does not offer at all. This is the only rate that is exclusive to H3 Max, and 480P is a drafting resolution, not a delivery one.

So the honest version: at matched resolution the two models cost the same, and H3 Max adds a cheaper draft tier below anything H3 has.

What 35x actually changes

The published claim is 35x the throughput of the official H3 endpoint, up to 50x at best. That number comes from the party that stands to gain from it, measured on their own benchmark, and should be read as a claim.

But it is a useful claim, because you can invert it. H3 Max returns a five-second 768P clip in under three seconds. If that is 35x the official endpoint's throughput, the same clip there is on the order of a minute and a half to two minutes. It is a derivation, not a stopwatch I held.

When it comes back fast enough, you stay in the loop. You watch the clip, change one clause, run it again. The prompt gets better because you are still holding the whole thing in your head.

When it takes a minute or more, you leave. You submit, switch tabs, come back, and by then you have lost the thread of what you were adjusting. So you compensate — you write longer prompts up front trying to get it right in one shot, and you batch variations you are not sure about because a round trip is expensive in attention.

That compensating behavior is the actual cost of slow generation, and it is invisible on any invoice. The clips are identical prices. The afternoons are not.

Twenty attempts, two tables

Take twenty attempts as a working assumption — swap in your own number and the ratio holds. Here is that job on both models, at 5 seconds each.

What it costs:

What it takes:

Same eight dollars at 768P. Half an hour against a minute.

If someone tells you H3 Max saves money, ask which resolution they were on before.

What you give up

2K. The one that decides it for a lot of people. If your deliverable is 2K, H3 Max cannot produce it at any price, and no amount of speed compensates.

The full reference set. H3 takes up to 12 files in one request: 9 images + 3 videos + 3 audio. H3 Max launched with text-to-video and image-to-video only and filled in reference-to-video days later. If your process depends on handing the model a reference video and asking it to match a performance, verify what the endpoint takes today.

A documented prompt ceiling. H3 states roughly 7,000 characters. The H3 Max schema does not publish one. A very long brief is something to test rather than assume.

A settled price. Both promotional rates expire this month, on different dates. Anything you cost out this week needs rechecking against $0.05, $0.08 and $0.13.

What H3 has no answer for

One capability only exists on the H3 Max side, and it is not a faster clip.

Director is a separate endpoint that holds a single continuous video stream open and maintains context across it, rather than chaining discrete clips. You can send new prompts mid-stream and the character, setting and continuity carry through. Sessions run to two minutes as standard.

Its meter is different from everything else here: it lists at $0.08 per second of generated video, and every session bills a minimum of 60 seconds regardless of how much of it anyone watched. The floor is a minute — around $4.80 at list price — so Director is not something you sample cheaply. There is no equivalent on H3 at all.

If what you want is interactive or audience-directed video, that is the endpoint, and it is worth reading what an hour of Director actually costs before you design around it.

Which one for which job

The combination that gets the most out of both: find the shot on H3 Max at 480P, deliver it on H3 at 2K. Prompts carry across, because it is the same model lineage. You pay $5 to find the shot instead of $13, and you still ship at full resolution.

Verdict

H3 Max is not an upgrade to MiniMax H3 and the name works against understanding that. It is a sibling with one thing traded for another: it gave up 2K and part of the reference set, and got back a generation time short enough that you never leave the loop.

At matched resolution it costs the same. The saving is 480P, which is a draft tier, and the real return is the half hour you get back on every shot that needs twenty attempts.

If you deliver at 2K, keep H3 in the pipeline and use H3 Max to find the shot. If you deliver at 768P, there is not much reason to wait on the slower endpoint. And if what caught your eye was the live-stream demos, that is Director rather than either of these — different endpoint, different meter, and a 60-second minimum that changes what you can afford to give away.

H3 runs in the browser here with free credits on signup, and our H3 Max page has the sample reel and the settings that differ. For longer pieces, the free H3 long video planner does the segment math before you spend anything.

More on MiniMax H3

• What is H3 Max — the speed, the Director meter, and what an hour of it costs
• What is MiniMax H3 — capabilities, pricing and honest limitations
• MiniMax H3 API guide — endpoints, input limits and the async workflow
• MiniMax H3 open weights — the release H3 Max was post-trained from
• MiniMax H3 prompt guide — how to structure prompts that actually control output
• MiniMax H3 vs Seedance 2.0 — the other comparison people ask for

This site is an independent third-party tool, not affiliated with MiniMax; credits here don't transfer.