Chinese AI firm MiniMax has launched H3-Image, an open-weight model capable of generating and editing images at 4K resolution. The model, which features 33.1 billion parameters and uses a diffusion transformer architecture, was released on November 10, 2026. It is available on Hugging Face and through the company’s APIs.
The model produces native 16-megapixel images, including 4096x4096 resolution, and supports wider formats up to 6272x2688. It requires approximately 6 sampling steps for generation and can process up to 16 reference images. Local deployment requires the additional Qwen3-VL-32B text encoder.
API pricing is set at approximately $0.022 for a 1K image and $0.069 for a 4K image, with promotional discounts available until November 17, 2026. ComfyUI has already integrated support for the new model, and a faster derivative version, P-H3-Image, was announced on November 3, 2026.
The community license restricts usage in the US, EU, UK, and South Korea. Furthermore, large corporations are required to obtain written authorization from MiniMax to utilize the model for their operations.
