MiniMax H3 lists at $0.13 per second through the Higgsfield API at 2K, which is $1.30 for a 10 second generation and $3.90 for a finished 30 second spot. That is the second highest per second rate in a catalog of 34 models, behind only Genjutsu at $0.154. On a subscription plan the same model is billed in credits, at a rate Higgsfield does not publish, so the dollar figure above is the API price and nothing else.
MiniMax H3 is the model you reach for when the shot needs a character, a motion reference and dialogue in one pass. Higgsfield's model guide describes it exactly that way: combine character images, motion references, and audio in one video, with dialogue and sound effects. It is not a model you iterate on. It is a model you point at a script.
If you are buying access to test it, referral signups currently take up to 50% off for three hours. The link is here, and we may earn a commission if you subscribe through it, at no extra cost to you.
The Published Rate: $0.13 a Second
The API pricing page lists MiniMax H3 at $0.13 per second, priced on a 2K configuration with a 10 second duration, across three modes: text to video, image to video and reference to video. It has been in the API catalog since July 1, 2026, and the page's own competitor comparison puts it at parity with fal at $0.13, which is to say Higgsfield is not undercutting anyone on this model.
Ten seconds, one generation: $1.30. Twenty clips of five seconds: $13. A hundred five second shots, which is a week of short form production: $65.
Where That Sits in the Catalog
The API rate card is the one place Higgsfield publishes per second prices openly, and it makes H3's position obvious.
| Model | List price per second | 30 seconds |
|---|---|---|
| Genjutsu | $0.154 | $4.62 |
| MiniMax H3 | $0.130 | $3.90 |
| Kling 4.0 | $0.126 | $3.78 |
| Seedance 2.5 | $0.118 | $3.54 |
| Seedance 2.0 | $0.080 | $2.40 |
| Genjutsu Light | $0.074 | $2.22 |
| Kling 3.0 | $0.071 | $2.13 |
| Kling 4.0 Flash | $0.063 | $1.89 |
| Kling 2.6 | $0.052 | $1.56 |
| Wan 3.0 Prime | $0.051 | $1.53 |
| Wan 3.0 | $0.038 | $1.14 |
Same 30 seconds, same output length, $1.14 on Wan 3.0 and $3.90 on MiniMax H3. That is 3.4x, and it buys you audio and dialogue, which Wan 3.0 does not generate in one pass. Pay for it only when the shot needs a voice.
What the Subscription Rate Card Does Publish
Here is the split that confuses people. The API has a price for H3. The subscription rate card does not list H3 at all. What it lists is the Hailuo family H3 sits on top of:
| Subscription line | Resolution | Credits per 6s | Credits per second |
|---|---|---|---|
| Minimax Hailuo 02 Fast | 512p | ~3 | ~0.5 |
| Minimax Hailuo 2.3 Fast | 768p | ~4 | ~0.67 |
| Minimax Hailuo 2.3 | 768p | ~6 | ~1 |
| Minimax Hailuo 2.3 Fast | 1080p | ~7 | ~1.17 |
| Minimax Hailuo 2.3 | 1080p | ~10 | ~1.67 |
Those are real published numbers on the pricing page as of November 18, 2026, and they give you a working estimate for H3 on a plan even without a line item for it. At roughly 1 credit a second on the 768p line, Starter's $0.070 credit implies about $0.07 a second, and the 1080p line at 1.67 credits a second implies $0.117.
Notice how close that $0.117 lands to the API's $0.13. That is not a coincidence worth ignoring: the API price is roughly what the same work costs inside a subscription, which means the API is not a discount channel for H3. It is the channel you use when you need concurrency, automation or a key instead of a browser tab.
A fair note on precision. H3 is newer than every Hailuo line in that table, and I have not seen its own credit cost, so treat the estimate as a bracket, not a rate. Generate one clip and read the balance delta before you build a budget on it. The same discipline applies to any model where the rate card goes quiet.
H3 Max Is a Different Purchase
MiniMax H3 Max arrived August 31, 2026 as MiniMax's post-trained variant of H3, tuned for prompt adherence and aesthetics. It caps at 768p, where the base H3 reaches 2K.
That cap inverts the usual assumption about the Max label. If your deliverable is 2K, Max cannot make it. Max is the version you pick when the prompt itself is the hard part, a specific action, a specific look, a specific expression, and you accept 768p in exchange for it landing closer to what you described. If the client wants resolution, base H3 is the only option on the board.
The verdict: MiniMax H3 at $0.13 a second is a dialogue model with a dialogue model's price tag, and it is worth it on the two or three shots per project that need lip movement and sound in the same pass. Draft those same shots on Wan 3.0 at $0.038 for timing and framing, then re-render the keepers on H3. For the model that H3 gets compared to most, Kling 3.0 cost per second is the other side of that decision, and the Seedance 2.5 breakdown covers the model with a published 35 credits per 5 seconds on subscription.