MiniMax rolled out its H3 video model as an open-weight release, but the license bars commercial use in the United States, United Kingdom, European Union and South Korea, and forces any company earning more than $20 million a year to seek written permission.

The headline-grabbing “open-weight” label suggests anyone can download, inspect and run the model, yet the fine print tells a different story. The restriction reminds us that open-weight does not automatically mean open-source, and it signals a shift in how AI firms package their most valuable assets.

Why the license matters

Open-weight means the model’s parameters are publicly available, so researchers can tinker with the core generator without relying on a hosted API. MiniMax’s H3 therefore looks like a generous contribution to the community. However, the commercial license imposes geographic limits and a revenue ceiling.

  • Territory block – Companies in the US, UK, EU or South Korea cannot use the model for any commercial product without a separate agreement.
  • Revenue trigger – Any organization that expects more than $20 million in annual sales must obtain written permission before deploying the model.

These clauses create a two-tier system: anyone can explore the model for free, but only a restricted set of businesses can monetize it without extra paperwork. Developers who assumed an open-weight release would be a plug-and-play foundation for commercial video-generation services now face a legal maze.

The emerging “open-weight, not open-source” pattern

MiniMax’s approach mirrors a growing trend in the AI industry:

  • Base model released – The core generator is downloadable, satisfying the research community’s appetite for transparency.
  • High-value components retained – Pre-processing pipelines, high-resolution upscaling tools and other performance-boosting modules stay on the company’s servers, often behind a paywall.

In practice, the open-weight H3 model can generate low-resolution video frames, but the polished output MiniMax showcases relies on proprietary services that are not part of the public release. Testing the released weights alone therefore gives an incomplete picture of the product’s capabilities.

This split-model strategy lets companies reap reputational benefits of open contributions while protecting revenue-generating features and navigating regulatory or export-control constraints. By limiting commercial use in key markets, MiniMax also sidesteps potential legal exposure in jurisdictions with stricter AI governance.

What developers should watch

  • Read the license before you download – The same place you see the parameter count should also display the usage terms. Ignoring the fine print can lead to costly compliance issues later.
  • Assess the missing pieces – If a model’s performance hinges on hosted services, factor the cost and latency of those services into any product roadmap.
  • Expect more of the same – As more firms balance openness with commercial protection, the “open-weight but not fully open-source” model is likely to become the norm rather than the exception.

MiniMax’s H3 release shows that the AI community’s push for transparency is now being tempered by business realities. The model’s public weights are a valuable research tool, but anyone hoping to build a commercial video-generation product must first navigate a patchwork of territorial and revenue-based restrictions.