MiniMaxImage modelActive
MiniMax H3-Image
Summary
MiniMax introduced H3-Image on November 11, 2026, as an image model with open weights. One checkpoint handles text-to-image and image editing and accepts up to 16 reference images. The model card describes about 16 megapixels; the API allows up to 20 million pixels and 1 to 10 output images per request.
The API model ID is `MiniMax-H3-Image`. It charges USD 2.30 per million text-input tokens, USD 1.50 per million image-input tokens, and USD 3.80 per million image-output tokens. See Prices for per-image examples and the time-limited 4K discount; image costs also vary with prompt length. The weights are under the MiniMax-H3-Image Community License Agreement. The optional `MiniMaxAI/MiniMax-H3-Image-PE` checkpoint expands image prompts; it is not an image generator.
Similar models
More models of the same kind as MiniMax H3-Image, chosen by model line, developer, size, price and release date.
- MiniMax Image-01MiniMax · same developer
- MOSS-Image-SFTOpenMOSS · released around the same time
- MOSS-Image-BaseOpenMOSS · released around the same time
- FLUX.2 Klein 9BBlack Forest Labs · released around the same time
- FLUX 3 Image KleinBlack Forest Labs · released around the same time
- GPT Image 2.5 SunburstOpenAI · released around the same time
- GPT Image 2.5 FlareOpenAI · released around the same time
- GPT Image 2OpenAI · released around the same time
All specifications
Architecture
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
- Shutdown
- None announced
- no source
API
Further details
- Image size
- The model card describes about 16 megapixels. The API accepts image areas from 655,360 to 20,000,000 pixels and up to 4K.
- Source: Image size
- Request limits
- 1 to 10 output images per request; prompts up to 32,000 characters. Without prompt enhancement, the API docs say only about the first 4,096 tokens take effect.
- Source: Request limits
- Output options
- JPEG, PNG, or WebP. The API does not support transparent backgrounds or masked editing.
- Source: Output options
- Weights
- The diffusion transformer has 33.1 billion parameters and BF16 weights. The Hugging Face repository is 68.65 GB; the reused H3 text encoder is in the separate H3 repository and is also required.
- Source: Weights
- Optional prompt expansion
- MiniMaxAI/MiniMax-H3-Image-PE is a separate optional text checkpoint (27B, FP4, 17.9 GB in the model zoo) that expands short image requests and recommends an aspect ratio. Its weights use the MiniMax-H3-Image Community License as well. For image editing, the release currently provides a prompt only, not PE-tuned weights.
- Source: Optional prompt expansion
Prices
- API token rates
- USD 2.30 per 1M text-input tokens, USD 1.50 per 1M image-input tokens, and USD 3.80 per 1M image-output tokens.Source
- Text-to-image example costs
- Example cost per image: USD 0.02179 at 1K, USD 0.02333 at 2K, and the regular USD 0.06879 at 4K. A 50% discount applies to a 4K image of at least 16,600,000 pixels until November 17, 2026, 16:00 UTC (exclusive); the example price becomes USD 0.03439. Costs vary with prompt-token count.Source
- Image-editing example costs
- Example cost per image at 1K/2K/4K output: with one 1K reference image, USD 0.02654/0.02658/0.07312; with three 1K reference images, USD 0.03563/0.04177/0.07914. A 50% discount applies to 4K images of at least 16,600,000 pixels until November 17, 2026, 16:00 UTC (exclusive), reducing those 4K examples to USD 0.03656 and USD 0.03957. These examples also vary with prompt-token count.Source