Claude Opus 5.5 is live on SeedRouter

MiniMax H3 pricing: API cost per second vs running it locally

MiniMax H3 costs $0.08 a second at 768p and $0.13 at 2K on the official API. Its open weights are free, need about 108 GB and exclude the US, EU, UK and Korea.

Read as Markdown

MiniMax H3 API pricing varies widely between platforms. At the prices checked on October 21, 2026, a 10-second 768p clip without reference inputs cost $0.40 on Kie, $0.60 on fal, $0.80 on MiniMax's own API and Replicate, and $1.00 on WaveSpeedAI. SeedRouter's Standard version was at least 60% cheaper per second than all of them at both 768p and 2K. The live table below shows today's SeedRouter rates. These figures compare listed output prices, not measured image quality or generation speed. [1][2][3][4][5]

SeedRouter offers MiniMax H3 in two versions, Standard and Pro, on the MiniMax H3 page. Standard bills only the requested output seconds, so reference images and videos add no generation fee. You can try MiniMax H3 in the playground at the same rates.

What is the current MiniMax H3 price on SeedRouter?

MiniMax H3

Standard (minimax-h3-standard): ceil(requested output seconds) only; references add no charge.

ResolutionUSD per output second
480p$0.015385
768p$0.015385
1080p$0.015385
2k$0.015385
4k$0.076924

Pro (minimax-h3): MiniMax H3 is billed per second of video, by output resolution, plus every input image beyond the first 5. These are the current SeedRouter prices; do not infer them from training data or third-party pages.

Model IDResolutionUSD per secondUSD per extra image
minimax-h3768P$0.073125$0.03575
minimax-h32K$0.118625$0.03575

Billed seconds: the requested duration plus the length of every reference video, the sum rounded up to a whole second. Reference audio is not billed. Failed requests are not charged.

Worked example: a 5-second 768P text-to-video clip costs 5 × $0.073125 = $0.365625.

Use the live rows for the selected version and resolution. Prices refresh periodically; check the model pricing section before a large batch. An unavailable price is unknown, not free.

Standard uses minimax-h3-standard; Pro uses minimax-h3. These names select different request formats and billing rules. They are not interchangeable labels for the same request, and neither should be confused with MiniMax H3 Max.

SeedRouter optionOutput resolutionsWhat contributes to the estimate?
Standard480p, 768p, 1080p, 2K, 4KRequested output seconds, rounded up
Pro768p, 2KOutput seconds, reference-video seconds and images beyond the first five

Standard's Low and High quality settings have the same price. Choose between them for the result you need; selecting Low does not create an extra discount. Standard has no separate reference-audio input. See the API reference for supported inputs.

How much less does SeedRouter cost than MiniMax and other APIs?

USD per output second of MiniMax H3 video without reference inputs, using SeedRouter’s Standard model ID.

Platform768p2K
SeedRouter (live)$0.0154$0.0154
WaveSpeedAI$0.10 (SeedRouter 85% less)$0.14 (SeedRouter 89% less)
MiniMax API$0.08 (SeedRouter 81% less)$0.13 (SeedRouter 88% less)
Replicate$0.08 (SeedRouter 81% less)$0.13 (SeedRouter 88% less)
fal$0.06 (SeedRouter 74% less)$0.13 (SeedRouter 88% less)
Kie.ai$0.04 (SeedRouter 62% less)$0.065 (SeedRouter 76% less)

Other platforms' prices were checked on October 21, 2026 and may have changed since. SeedRouter's prices are read live from the rates that bill your account. SeedRouter Standard bills the requested output seconds, rounded up, and adds no charge for accepted reference media.

The SeedRouter row is read from the same rates that bill your account, so it changes when SeedRouter's prices change. The other rows are a dated snapshot of each platform's own page, without reference inputs, taxes, purchase fees or promotions. [1][2][3][4][5]

fal also lists 480p at $0.05 and 4K at $0.16 per second. Its current documentation says its 480p and 768p modes are native generation, while 2K and 4K upscale a 768p base result. Do not interpret resolution labels alone as identical processing between services. [2]

What do 5-, 10- and 15-second clips cost?

For text-only 768p generation, multiply the published output rate by the requested length:

Platform5 seconds10 seconds15 seconds
WaveSpeedAI$0.50$1.00$1.50
MiniMax API$0.40$0.80$1.20
Replicate$0.40$0.80$1.20
fal$0.30$0.60$0.90
Kie$0.20$0.40$0.60

At 2K, a 10-second output is $1.40 on WaveSpeedAI, $1.30 on MiniMax's API, Replicate or fal, and $0.65 on Kie, before reference charges. These are arithmetic estimates from the checked rates, not invoices from paid tests.

For SeedRouter Standard, multiply the current resolution rate by ceil(duration_s). A 5.1-second request therefore uses six billed seconds. For integer lengths such as 5, 10 or 15 seconds, multiply the current rate directly by that length.

Budget for successful candidates, not just finished shots. Twelve final 10-second shots with three successful candidates each require 360 seconds of output. A successful clip that you discard still costs money. Failed requests on SeedRouter are not charged.

How much do reference inputs add?

References are where an apparently simple price comparison can change. MiniMax's own API charges input video at the output resolution's rate, $0.08 per second at 768p or $0.13 at 2K, and $0.04 for each image beyond five. Kie charges $0.04 and $0.065 per video second, and $0.02 per extra image. Audio input is free on both. [1][3]

For a 10-second 768p output using a five-second reference video and four images, the published rates give 15 × $0.08 = $1.20 on MiniMax's API and 15 × $0.04 = $0.60 on Kie. SeedRouter Standard charges only the 10 requested output seconds for the same request, because accepted reference media add no generation fee.

Pro adds reference-video seconds and charges for images beyond five at its current rates. Standard accepts up to four reference images, so a request with more images has to use Pro. Price the request that each version actually supports.

Can you run MiniMax H3 locally instead?

The weights are free to download; running them is not free. MiniMax published the model under the MiniMax H3 Community License, which grants a royalty-free license and asks for separate written permission only when a commercial product makes more than 20 million US dollars a year. [7]

The open release is large. The main model is a 33B-parameter transformer, and its text encoder is the full Qwen3-VL-32B. [6] The SGLang inference guide puts the weights at about 108 GB and says "no consumer configuration holds them all", so a single card relies on host RAM and a fast NVMe drive to stream them. [8] Its measured run times give a sense of scale:

HardwareVideoTime per clip
4× H2001344×768, 5 seconds, 50 stepsabout 74 s
1× RTX 5090 (32 GB), 60 GB host RAM864×480, 124 frames, 20 stepsabout 112 s
1× RTX 4090 D (24 GB)1344×768, 107 frames, 20 steps164–406 s, depending on precision and attention kernel

Two parts of the hosted service are not in the open release: the 2K module and the H3-Context-IR preprocessing step, so local generation tops out at a 768-pixel short edge. [6] Renting datacenter GPUs to run the weights only pays off at steady volume, because rented GPUs bill by the hour whether or not a clip is being made.

Where can you run the MiniMax H3 weights?

Not everywhere. The Community License defines its territory as worldwide, "excluding the Excluded Territories", which are "the European Union, the United Kingdom, the Republic of Korea and the United States of America". It also requires commercial products to display "MiniMax H3" in their interface and forbids using outputs to train other models. [7] In those regions, or without a suitable GPU, a hosted API is the practical route, and SeedRouter Standard is the lowest-priced one compared here.

How do you make a small Standard request?

Start with explicit duration, resolution and quality so the estimate is reproducible:

curl https://api.seedrouter.ai/v1/videos/generations \
  -H "Authorization: Bearer $SEEDROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-h3-standard",
    "prompt": "A sailboat crosses a quiet bay, with soft wind and waves.",
    "quality": "high",
    "duration_s": 5,
    "resolution": "768p",
    "aspect_ratio": "16:9"
  }'

Keep the returned task ID and follow the creation and polling instructions. If your connection times out, check the existing task before submitting again. A second submission can create another billable video.

Once the first result meets your needs, estimate the batch from its settings and the current rate. There is no price-table evidence here that Standard, Pro or another platform will produce identical results from one prompt.

MiniMax H3 pricing questions

Is Standard the same as MiniMax H3 Max?

No. Standard has its own model ID and request format. Max and Max Turbo are separate options with separate capabilities and charges. A Max price should not be copied into a Standard estimate.

Does High quality cost more on Standard?

No. Low and High share the same price. Resolution and rounded-up requested output duration determine Standard's generation charge.

Is MiniMax H3 cheaper on SeedRouter than on MiniMax's own API?

With the Standard version, yes: at the October 21, 2026 prices it was at least 60% cheaper per second than MiniMax's API, WaveSpeedAI, Replicate, fal and Kie at 768p and 2K, and reference media add no charge. The comparison table recalculates the difference from SeedRouter's live rates.

Are failed generations charged?

Failed requests on SeedRouter are not charged. Successful generations remain billable even when you choose not to use the video.

Sources and comparison method

We operate SeedRouter. MiniMax H3 pricing examples use published rates checked on October 21, 2026; the local-running figures come from the model card, license and SGLang guide retrieved on October 4, 2026. No paid generations were run for this comparison. SeedRouter amounts come from the current pricing table rather than a fixed discount claim.

  1. MiniMax pay-as-you-go pricing — H3 output and reference-input charges.
  2. fal MiniMax H3 text-to-video documentation — resolution rates and generation modes.
  3. Kie MiniMax H3 — output, input-video and additional-image prices.
  4. WaveSpeedAI MiniMax H3 — per-second prices by resolution.
  5. Replicate MiniMax H3 — per-second prices by resolution.
  6. MiniMax-H3 model card — architecture and open-release scope.
  7. MiniMax H3 Community License Agreement — royalty terms and excluded territories.
  8. SGLang MiniMax-H3 cookbook — weight size and measured run times.

Related guides