- MiniMax H3 pricing starts at $0.09 per second for listed 768p output.
- 2K generation is listed at $0.13 per second, making a 15-second clip $1.95.
- Reference images are free for the first five, then cost $0.04 each.
- Reference video usage is billed by input duration at the selected output rate.
- API status is live in 2026, while the promised model weights are not yet available here.
MiniMax H3 pricing at a glance
MiniMax H3 is a hosted multimodal AI video model rather than a conventional game or entertainment title. Its pricing is based mainly on generated video duration and output resolution. The API accepts text, images, video, and audio references, allowing creators to plan short-form production around a per-second budget.
Video Highlights:
- Native stereo sound is included in the demonstrated 2K video workflow.
- The documented design supports outputs of roughly 4 to 15 seconds through the create endpoint.
- H3 combines multiple reference types, but first-and-last-frame generation uses a separate mode.
- The launch examples focus on film openings, product pages, posters, and fashion advertising.
| Output option | Listed rate | 15-second output | Best use |
|---|---|---|---|
| 768p | $0.09/second | $1.35 | Drafts and rapid iteration |
| 2K | $0.13/second | $1.95 | Final-quality short clips |
| Extra reference images | $0.04 each after five | Depends on count | Character or brand consistency |
| Reference audio | No listed charge | No listed charge | Voice, music, or sound direction |
The most important distinction is between output cost and input reference cost. A simple text-to-video request can remain close to the base rate. A heavily directed shot with many images and video references may cost more than the generated clip alone.
Start creative exploration at the lower listed resolution when available, then reserve 2K generations for approved shots. This separates concept testing from final production spending.
How to calculate MiniMax H3 pricing
The basic calculation is straightforward: multiply the output duration by the selected per-second rate. At the listed rates, every additional second adds $0.09 for 768p or $0.13 for 2K. This makes duration control one of the easiest ways to manage a project budget.
| Clip length | 768p estimate | 2K estimate | Planning note |
|---|---|---|---|
| 5 seconds | $0.45 | $0.65 | Useful for quick tests |
| 10 seconds | $0.90 | $1.30 | Good for short social shots |
| 15 seconds | $1.35 | $1.95 | Maximum length in the cited workflow |
Reference images change the estimate after the first five. For example, a request with eight images adds three chargeable images, or $0.12, before any reference video input is counted. Reference audio is listed as free, but reference video is billed according to its input duration and selected output rate.
A practical estimate can use this formula:
Estimated cost = output seconds × output rate + chargeable image references × $0.04 + reference video input cost
The available documentation has some launch-period differences. One model overview describes clips from 5 to 15 seconds, while the create endpoint lists 4 to 15 seconds. The pricing page mentions 768p, but the current create endpoint described in the available material exposes 2K as the active resolution. Treat the operational endpoint as the safer reference when building an integration.
Pricing and endpoint details can move quickly during a launch. Confirm the live API response, supported resolution, duration limits, and billing behavior before committing to a production budget.
API setup and reference modes
MiniMax H3 requires a text prompt, then allows creators to add supporting references. The multimodal workflow is designed for shots that need a consistent subject, camera language, performance, or soundtrack. It is more suitable for controlled production than for treating every generation as an isolated prompt.
Define the shot
Write the text prompt first. Specify the subject, action, framing, camera movement, lighting, style, and sound direction before adding references.
Choose the reference mode
Use reference images, videos, and audio when identity or creative continuity matters. Use first-frame, last-frame, or both when controlling the shot boundaries is more important.
Stay within input limits
The documented reference workflow supports up to nine images, three videos with a combined duration of 15 seconds, and three audio references.
Estimate the request
Calculate output seconds first, then add chargeable image references and the billed duration of any reference videos.
Test before scaling
Generate a small set of representative shots and review identity preservation, motion, audio alignment, text rendering, and consistency before ordering many clips.
The two reference approaches should not be mixed casually. First-and-last-frame generation cannot be combined with the reference image, reference video, and reference audio mode described for omni reference input. Select the workflow that matches the production problem instead of trying to maximize every input field.
Omni References
Combine images, video, and audio to guide identity, movement, performance, and sound in one request.
Frame Control
Use a first frame, last frame, or both when the start and end composition matter most.
Commercial Direction
Plan prompts around products, titles, posters, and advertisements where legibility and art direction are important.
More references do not automatically mean a better result. Use only assets that clarify the intended shot, because unnecessary input can increase cost and complicate troubleshooting.
Value, quality, and production fit
The strongest value proposition is the combination of multimodal references, 2K output, native audio, and competitive per-second pricing. H3 is particularly interesting for creators who need a hosted API and want to produce directed clips rather than generic text-to-video scenes.
An independent arena signal cited for the launch placed H3 second in text-to-video with audio, with an Elo score of 1242 from just over 5,000 samples. Gemini Omni Flash was listed first at 1246. The 95% confidence ranges overlap, so this result indicates near parity in that arena rather than a decisive lead.
| Evaluation area | Reported position | Editorial reading |
|---|---|---|
| Text-to-video with audio | 2nd | Competitive result with overlapping confidence ranges |
| Video editing with audio | 1st claim | Treat as a company-reported ranking until independently checked |
| Image-to-video without audio | 2nd claim | Promising, but broader testing is still useful |
| Open-weight readiness | Pending | Repository, license, checkpoints, and hardware guidance remain important |
The official launch examples demonstrate ambitious commercial categories, including a film opening with titles, a stylized product website, a vertical animated poster, and an e-commerce fashion advertisement. These examples are valuable because they test text rendering, composition, branding, and art direction more severely than a generic landscape prompt.
They are still selected demonstrations. They do not establish generation success rates, rejected outputs, latency, or long-campaign identity consistency. Teams should evaluate those factors with their own prompts and acceptance criteria.
Before approving a production budget:
- Confirm the live endpoint resolution and duration limits
- Separate draft generations from final 2K outputs
- Count chargeable image and reference video inputs
- Test identity, motion, audio, and text consistency
- Review the API terms and current billing documentation
MiniMax H3 is a strong candidate for teams that need hosted generation, directed references, short 2K clips, and predictable per-second estimation.
MiniMax H3 pricing FAQ
Q: How much does a 15-second MiniMax H3 clip cost?
The listed estimate is $1.35 at 768p or $1.95 at 2K before applicable reference-input charges.
Q: Are reference images included in MiniMax H3 pricing?
The first five reference images are listed as free. Additional images are priced at $0.04 each, while reference video is billed by input duration at the selected output rate.
Q: Is MiniMax H3 available as downloadable model weights?
The API is live in the cited 2026 launch material, but the promised weights are described as coming later and are not presented as an available self-hosting release.
Q: Which MiniMax H3 workflow should creators use?
Use omni references for identity, camera, performance, and audio continuity. Use first-frame or last-frame controls when shot boundaries are the primary requirement.
Use the listed per-second rates as a starting estimate, then validate live endpoint behavior and reference billing before scaling beyond test generations.