01
Faster model
FastH3 needs only a few denoising steps, so a 10-second clip with audio renders in about 6 seconds on eight GB200 GPUs. More video per GPU-hour lowers the cost per attempt.
Blog · October 8, 2026
MiniMax H3 API price vs self-hosting: the real cost per clip
Teams compare video models by price per second. In production, the number that matters is cost per accepted clip: what you pay, including retries and rejects, for each clip that passes review. A dedicated, customized H3 deployment lowers both sides of that ratio.
The metric
Cost per accepted clip
Cost per accepted clip = total production cost ÷ clips that pass review.
Total cost includes every generated attempt, retries, review time and idle capacity. If only half of your clips pass review, your real cost per clip is double the list price. A lower price per second does not help if the acceptance rate falls.
List prices
MiniMax H3 and Seedance 2.0 API price per second
| Model and tier | List price | 10-second clip |
|---|---|---|
| MiniMax-H3, 768P | $0.08 / s | $0.80 |
| MiniMax-H3, 2K | $0.13 / s | $1.30 |
| Seedance 2.0, 720p | $0.15 / s | $1.50 |
| Seedance 2.5, 720p | $0.231 / s | $2.31 |
Sources: MiniMax pricing and BytePlus pricing, read October 8, 2026. These are list prices for different models and resolutions, so they do not compare quality.
Dedicated GPU cost
Self-hosting cost: GPU cost by utilization
| Utilization | GPU cost per second of 768p video |
|---|---|
| 100% | ≈$0.0144 |
| 50% | ≈$0.0288 |
| 25% | ≈$0.0576 |
Illustrative case: eight RTX PRO 6000 GPUs at $3 per GPU-hour, or $576 per day, and about 4,000 ten-second 768p clips with audio per day at full use. It excludes service fees, storage, network, review and rejected clips. A fixed cost is cheap only when the host stays busy.
Calculator
API vs self-hosting cost calculator
Defaults: the MiniMax-H3 768P list price and the illustrative eight-GPU host. The host cost here is GPU cost only. Change any value.
Where Nuva Lab lowers the cost
Four levers on cost per accepted clip
02
Higher acceptance
A model customized on your brand and references aims to get more clips through review on the first try. We measure this on your own briefs. How customization works.
03
Fewer manual steps
Agents run batches, retries and checks, so your team reviews results instead of operating tools. Agents and the feedback loop.
04
Right deployment
A dedicated host suits steady, high volume. At low or spiky volume the API is often cheaper, and we will tell you so.
FAQ
Questions
How much does the MiniMax H3 API cost?
The MiniMax list price is $0.08 per second at 768P and $0.13 per second at 2K, read October 8, 2026. A 10-second 768P clip costs $0.80 per attempt.
How much does Seedance 2.0 cost per second?
The BytePlus list price for Seedance 2.0 at 720p is $0.15 per second, read October 8, 2026. A 10-second clip costs $1.50 per attempt.
Is a dedicated host cheaper than the MiniMax H3 API?
Only at steady, high volume. At $576 per day of GPU cost, a host breaks even with the $0.08-per-second API at about 720 ten-second clips per day, before review and rejected clips. Below that, the API is usually cheaper.
What is cost per accepted clip?
It is the total production cost divided by the clips that pass review. It includes retries, rejected clips, review time and idle capacity, so it shows the real cost of usable output.
Why not compare price per second only?
Price per second ignores acceptance rate. If a cheaper model needs twice as many attempts to get a usable clip, it costs more per accepted clip.
Sources and further reading
Read next
Find your cost per accepted clip
Tell us your volume and review process. We'll size a dedicated H3 or FastH3 deployment and compare it with your current API cost.
Get started · 1 min