MiniMax H3 Tops the Arena — And Has No Price (Aug 2026)

MiniMax H3 launched on July 31, 2026, and within 48 hours it sat at arena rank #1 across 2 of the leaderboards we track, on the strength of 36 weighted mentions in our launch-week signal. It is, by momentum, one of the two biggest AI-video stories of the week. And we cannot tell you what it costs to call — because there is no verified API price for it in our dataset at all. Not a high one. Not a placeholder. Zero providers we monitor sell it as a callable endpoint yet.

That gap between "everyone is talking about it" and "nobody can price it" is the whole story, and it is not an accident. This page explains exactly what is knowable about MiniMax H3 today, why the number is missing, and — since you probably came here to spend money, not admire a leaderboard — what you can actually pay this week instead. The live standing stays at /models; the moment a real H3 rate appears, it lands there normalized to USD-per-second.

Everything below is checked against our dataset as of August 2, 2026. Where we have no number, we say so rather than inventing one.

What launched: the verifiable facts

Here is what our launch-week signal actually captured, stripped of hype. MiniMax H3 is MiniMax's new omni-modal video model, and the headline claims traveling with it — repeated across the coverage we harvested — are consistent: 15-second clips, 2K resolution, native stereo audio, generated in one pass. The launch titles in our signal read verbatim:

• "MiniMax Releases MiniMax H3: An Omni-Modal Video Model That Generates 15-Second 2K Clips With Native Stereo Audio"
• "MiniMax H3: 2K AI Video With Native Audio and Open Weights"
• "MiniMax H3 Tops Video Editing Leaderboard, Plans Open Weights Release"

That last one matters more than the specs. H3 is going open-weights. That single decision explains why this section is short on dollar signs: a model shipped as open weights leads with a license and a download, not a hosted rate card. The pricing follows later, if and when third-party hosts stand up inference endpoints and publish per-second rates. Until one does, there is nothing to quote.

For context on who is shipping it, MiniMax is the same developer behind Hailuo, which we do track and price — Hailuo 2.3 runs a verified $0.0267/s at 720p on PiAPI. So this is not an unknown lab; it is an established player that chose to open-source its frontier model rather than gate it behind an API meter on day one.

Why there is no price (and why that's normal)

Our system has a hard rule for the "available now" ranking: a model only counts as broadly available when it clears a gate — either ≥2 independent API providers with verified pricing, or ≥2 consumer platforms offering it without a waitlist. MiniMax H3 clears neither. In the standing data it shows 0 verified providers and 0 clean platforms, which is why it does not appear in the available-now table at all. It ranks only in the frontier lens — the one that deliberately includes early-access and not-yet-purchasable models on momentum alone.

This is exactly the pattern we saw with its same-day twin. Seedance 2.5 also launched July 31, 2026 — the two shipped within hours of each other, and the coverage framed it as a duel: "China's AI Video Battle Picks Sides: Seedance 2.5 Stays Closed, MiniMax H3 Goes Open." ByteDance kept Seedance 2.5 closed and published a first-party rate immediately. MiniMax did the opposite: open the weights, skip the rate card. Two philosophies, same launch day, and only one of them gives you a number to put in a budget.

So the absence of an H3 price is not us failing to find one. It is the direct, predictable consequence of an open-weights launch. Anyone quoting you a firm "$X per second for MiniMax H3" this week is quoting a number that does not yet exist in any verifiable form. Treat it the way you would any un-sourced price: don't.

What Seedance 2.5 costs — the only launch with a real number

If your actual question is "which of this week's new models can I pay for right now," the answer is the closed one. Seedance 2.5's first-party host, BytePlus ModelArk, published verified text-to-video pricing on launch:

Source: BytePlus ModelArk, verified August 1, 2026. On a 5-second clip that is $0.103/s × 5 = $0.515 at 480p and $0.231/s × 5 = $1.155 at 720p (BytePlus's own per-video cross-check lists $0.514 and $1.156, matching to the cent). That is the only verified per-second rate attached to any July 31 launch. If you need a new-generation model callable this week with a knowable cost, it is Seedance 2.5 or nothing — and we break down that launch in full at /models/seedance-2.5. H3, for now, is a leaderboard entry you cannot invoice.

What to actually pay this week

Chasing the un-purchasable model is how you burn a week waiting on a waitlist. Here is what the same money buys today, using only rates we have marked verified — cheapest per second, model by model, from our tracker:

Every rate above is a verified row in our dataset, not a "reported" one — the distinction matters, because there are cheaper reported numbers floating around (a Replicate Seedance 2.0 480p listing at $0.08/s, for instance) that we have not independently confirmed, so we don't lead with them.

The practical read comes down to three routes, each priced off exactly one model:

• Workhorse — Seedance 2.0 at a verified $0.10/s. Frontier quality with native audio at a real rate. The per-clip math is dead simple: an 8-second clip is $0.10/s × 8 = $0.80.
• Cheapest verified — LTX-2 at $0.016/s. The lowest confirmed rate on the whole board. That same 8-second clip is $0.016/s × 8 = $0.128.
• MiniMax, but payable today — Hailuo 2.3 at $0.0267/s. If you specifically want a model from H3's own lab, this is the one you can actually invoice, from the maker that just open-sourced H3.

Run your real clip volume through the calculator — it prices your workload live against every verified provider we track, so you can see the actual dollar gap between, say, a thousand LTX-2 clips and a thousand Seedance 2.0 clips before you commit a key.

When H3 gets a price, you'll see it here first

The instant a provider stands up MiniMax H3 as a callable endpoint and publishes a per-second rate, it enters our dataset, gets normalized to USD-per-second, and appears in the /models catalog with a verification date attached — so you know it's real, not launch-day vapor. Given the open-weights release, expect the first hosted rates from the usual aggregators, not a first-party meter.

We're watching two things specifically: how far below Seedance 2.5's $0.103/s the first H3 host lands (open models with strong arena numbers get undercut fast once providers compete), and whether H3 shows up on the consumer platforms we track for deals, where a flat-monthly "unlimited" offer would appear first.

Until then, the honest bottom line stands: MiniMax H3 is the model everyone is ranking and nobody can price. If you need pixels on a timeline this week, buy a model with a verified rate — start at /models for the full priced catalog, or size your spend at /calculator. Don't pay for a leaderboard.

Standing, launch dates, and pricing verified against the GenRates dataset on August 2, 2026. MiniMax H3 launched July 31, 2026 with no verified API pricing in our data as of this writing; figures for other models are verified rows unless noted as reported. A model version is a distinct product — H3's eventual price will not be Hailuo's or Seedance's. Rates change fast, and open-weights pricing especially — check the live tracker before committing. Some links on this site may be referral links.