Hailuo AI pricing: 2.3 API costs and MiniMax H3 update
Quick Answer
Quick answer
MiniMax still lists Hailuo 2.3 at $0.28 for a six-second 768p API generation and Hailuo 2.3 Fast at $0.19 for the same format. MiniMax H3 is now the current open-weight family and has a different per-second price card. Consumer Hailuo subscriptions, MiniMax's direct API, and fal's separate H3 Max variant are distinct purchases. Prices checked September 13, 2026 against the linked provider pages. See MiniMax's Hailuo pay-as-you-go table.
Direct API cost per generation
Model | 768p, 6 seconds | 768p, 10 seconds | 1080p, 6 seconds |
|---|---|---|---|
$0.19 | $0.32 | $0.33 | |
$0.28 | $0.56 | $0.49 | |
$0.28 | $0.56 | $0.49 |
These are the published combinations in the pricing reference. Do not infer an unlisted 1080p ten-second option or apply one model’s rate to a different endpoint.
MiniMax H3 is a separate current model family
The current MiniMax H3 pricing card lists generated video at $0.08 per second for 768p and $0.13 per second for 2K. Reference audio is free; the first five reference images are free and additional images are $0.04 each. Reference video is billed by its input duration at the selected output resolution. These rates do not replace the Hailuo 2.3 rows above; they apply to a different model family and billing contract.
MiniMax H3 item | Current direct-API price | Practical example |
|---|---|---|
Generated video at 768p | $0.08 per output second | 10 seconds: $0.80 before retries |
Generated video at 2K | $0.13 per output second | 10 seconds: $1.30 before retries |
Reference audio | Free | Confirm duration and endpoint limits |
Reference images | First five free; then $0.04 each | Eight images add $0.12 |
Reference video | Billed by input duration at the selected output resolution | Trim the reference to the required segment |
Do not confuse base MiniMax H3 with H3 Max, a separately named fal-hosted, fal-post-trained variant with its own endpoints and prices. Record provider, endpoint ID, model variant, resolution, reference inputs, promotional terms, and retrieval date for every cost comparison.
What does a finished set of clips cost?
For twenty accepted six-second 768p clips, allowing three attempts per accepted clip means sixty generations. At $0.19 per attempt, the Fast example costs $11.40; at $0.28, the Hailuo 2.3 example costs $16.80.
The three-attempt allowance is hypothetical. These totals exclude taxes, other services, storage, and editing. Run a small representative project to learn your own retry rate before forecasting production spend.
API pricing versus Hailuo subscriptions
A consumer plan may offer credits, queues, feature access, and promotional allowances. Those credits do not automatically fund MiniMax’s direct API. Conversely, an API rate does not establish what a consumer subscriber pays for each finished video.
The older approximate subscription table has been removed because it combined uncertain plan figures and generalized usage estimates. Confirm the current Hailuo offer in its own checkout if you prefer a browser subscription.
Which model should you choose?
Choose the lowest-cost supported option that produces acceptable results on your actual scenes. Compare motion, subject continuity, required resolution, and available input modes. The word “Fast” is a model label, not a guarantee of end-to-end completion time under every queue condition.
Comparing another provider
A host can expose Hailuo alongside other models while using its own pricing and limits. Record the provider as well as the model. Magic Hour’s image-to-video workflow is one place to compare a generation task within a broader media workflow; use the displayed model-specific quote, not a generic credit conversion.
Where should you compare current MiniMax models?
Use the open-weight video model guide for MiniMax H3 deployment and licensing context. Use the AI video pricing index to compare model and platform costs with the same duration, resolution, audio, references, and retry assumptions. A cheaper request is not necessarily a cheaper accepted clip.
Cloud shadows move slowly across the scene while foreground plants respond to a light breeze. The camera makes one gentle push forward. Preserve the original composition, subjects, and visual style. One continuous shot.
Compare an image-to-video job
Use the same source image and brief, review the displayed model and credit estimate, and compare completed results and retries rather than vendor credit counts.
Try Image-to-VideoFrequently asked questions
For twenty accepted six-second 768p clips, allowing three attempts per accepted clip means sixty generations. At $0.19 per attempt, the Fast example costs $11.40; at $0.28, the Hailuo 2.3 example costs $16.80.
The three-attempt allowance is hypothetical. These totals exclude taxes, other services, storage, and editing. Run a small representative project to learn your own retry rate before forecasting production spend.
Choose the lowest-cost supported option that produces acceptable results on your actual scenes. Compare motion, subject continuity, required resolution, and available input modes. The word “Fast” is a model label, not a guarantee of end-to-end completion time under every queue condition.
Use the open-weight video model guide for MiniMax H3 deployment and licensing context. Use the AI video pricing index to compare model and platform costs with the same duration, resolution, audio, references, and retry assumptions. A cheaper request is not necessarily a cheaper accepted clip.
Runbo Li is the Co-founder and CEO of Magic Hour, where he builds AI video and image tools for content creation. He is a Y Combinator W24 founder and former Data Scientist at Meta, where he worked on 0-1 consumer social products in New Product Experimentation. He writes about AI video generation, AI image creation, creative workflows, and creator tools.
