China's MiniMax Locks Americans Out of Its 'Open' Image Model
MiniMax-H3-Image is topping Hugging Face's trending list and beating every open rival on image-editing scores, but the license bars self-hosting in the US, UK, EU and South Korea -- the exact markets closed labs serve.

MiniMax calls its newest release an open-weight model. Try reading the license before you celebrate.
MiniMax-H3-Image landed on Hugging Face on Tuesday, Nov. 10, and developers noticed immediately: ComfyUI shipped day-zero support, vLLM-Omni shipped day-zero support, and ten different quantized GGUF builds showed up within 72 hours. The file is real. The permission to run that file on your own hardware inside the United States, the United Kingdom, the European Union or South Korea is not.
(cite index="20-6,20-8">The license defines the "Applicable Territory" as worldwide, excluding the "Excluded Territories," which it names as the European Union, the United Kingdom, the Republic of Korea and the United States of America. (cite index="20-1,20-2">You may not use, reproduce, modify, distribute, or display the works or their outputs outside that territory. Four of the world's richest AI markets carved out of a release everyone calls open source.
The File Is Real, the Benchmarks Are Decent
(cite index="1-29">The diffusion transformer has 33.1 billion parameters, a dense single-stream design with 50 layers, hidden size 5,376, 56 attention heads and 3D multimodal rotary position embeddings. (cite index="1-10">It generates images of about 16 megapixels at any aspect ratio, from 4096x4096 up to 6272x2688, directly and without an upscaler.
Four days after weights went up, Artificial Analysis had numbers. (cite index="11-7">On the AA-Image-Editing v2.1 leaderboard it scored 1,138 Elo over 3,137 sampled comparisons, first among open-weight entries. (cite index="11-8">On text-to-image it placed twenty-fifth overall, first among downloadable models. Editing is where it leads the open field; generation ties Alibaba's Qwen-Image-2.1.
(cite index="12-7,12-8">A 1K text-to-image render costs about $0.022 on the Standard API tier, with a 50 percent discount on 4K output running until Nov. 17, 2026. (cite index="17-4">MiniMax is pitching a 16-megapixel image in ten seconds at 50 to 70 percent the cost of mainstream tools. That is the same undercutting playbook the lab ran on its M3.1 Flash text model in October, now compressed to zero days between hosted launch and open weights instead of four.

Open Except Where the Money Is
Here is the part the trending chart doesn't show. (cite index="22-13">The weights are public, but the Community License excludes the US, EU, UK and South Korea, requires written authorization above $1 million in annual revenue or for hosted services, and forbids training other models on its outputs.
MiniMax used nearly identical language on its H3 video model three months earlier. When pressed on it, (cite index="27-8">the company made clear a developer in San Francisco, London, Berlin or Seoul can use MiniMax's hosted API or consumer-facing Hailuo product freely; the exclusion covers only downloading the checkpoint and running it on your own infrastructure inside those borders. A MiniMax representative tied the carve-out to (cite index="26-9">the company's ongoing copyright lawsuit with Disney, Universal, and Warner Bros. Discovery over its Hailuo AI video-generation service, a case that (cite index="26-63">a court has allowed to proceed but has not resolved.
Translation: the weights are bait for developer goodwill and Hugging Face trending points. The regions that would actually pay enterprise rates get steered straight to the metered API, where MiniMax controls every token and every dollar.
A 135-gigabyte download you cannot legally run at home is not open source. It is a very effective billboard.
What to watch: whether MiniMax's API traffic spikes from US and EU developers who downloaded the weights, read the license, and routed to the paid endpoint instead; whether the next Chinese lab to trend on Hugging Face ships the identical four-country carve-out, turning a one-off into the industry's quiet standard.
- MiniMaxAI/MiniMax-H3-Image · Hugging Face
- MiniMax-H3-Image Tops Both Open-Weights Image Boards
- MiniMax releases open-weight H3-Image model with 4K output on Hugging Face
- Every MiniMax AI Model Explained and Compared (Nov, 2026)
- LICENSE · MiniMaxAI/MiniMax-H3-Image at main
- MiniMax H3-Image: 4K Open Weights, Price and License
- MiniMax-H3-Image on Mac: 135GB in BF16, Not Licensed in the US
- MiniMax H3 Explained: Inside the 33B Omni-Transformer You Can't Legally Self-Host in the US
- Three-Way AI Image Generation Battle: MiniMax Enters on Speed and Low Price
- Abiray/MiniMax-H3-Image-GGUF · Hugging Face
- MiniMax-H3-Image · Hugging Face
- Hearthscale/MiniMax-H3-Image-GGUF · Hugging Face
- MiniMaxAI/MiniMax-H3-Image at main
- MiniMax H3-Image: Bringing 4K Image Generation into the 10-Second Era - MiniMax Research
- GitHub - MiniMax-AI/MiniMax-H3-Image: MiniMax-H3-Image, an open-weight model for 16-megapixel image generation and editing. · GitHub
- Comfy-Org/MiniMax-H3-Image at main
- akhaliq/minimax-h3-image-pdmd-2step · Hugging Face
- MiniMax H3-Image: Bringing 4K Image Generation into the…
- MiniMax Drops MiniMax-H3-Image, a 33B Open Model Generating 16MP Photos in 6 Steps
- MiniMax-H3-Image: Three Launch Claims, Three Paper Trails
- MiniMax H3 Image is here. How to use MiniMax H3 Image for free?
- MiniMax-H3-Image: 16 MP in 10 Seconds, Still Unmeasured
- MiniMax H3-Image: an Open Image Model That Takes 16 References
- omsi/sources/huggingface.co/minimax-h3-license/2026-10-26.md at master · wsxiaoys/omsi · GitHub
- MiniMax H3: the weights are out, the licence excludes the European Union — CSIDIA
- MiniMax H3-Image: Price, 135 GB Download and Licence Limits
- Why MiniMax H3’s Open-Weight License Excludes the U.S., EU, UK, and South Korea