Course contentsShow
← AI Engineering
Lesson 1065 of 1,88626. Self-Hosted LLM DeploymentFree lesson

Model Families and Licensing

Survey popular open-weight models (Llama, Mistral, Falcon) with their licenses and commercial use terms.

Model Families and Licensing

What you'll learn: How to navigate the landscape of popular open-weight models and understand their licensing implications for commercial deployment.

Understanding Open-Weight Model Families

When deploying self-hosted LLMs, you're not just choosing based on performance—you're also selecting a license that determines how you can legally use the model. Unlike traditional open-source software with standard licenses (MIT, Apache 2.0), many AI models use custom licenses with specific restrictions.

Think of model licenses like rental agreements: some let you use the space for any purpose (commercial-friendly), while others restrict business use or require you to share modifications (copyleft-style).

Major Model Families and Their Licenses

Llama (Meta)

Meta's Llama 2 and Llama 3 families use the Llama Community License. You can use them commercially if your service has fewer than 700 million monthly active users. Above that threshold, you need Meta's permission. The weights are "open" but not truly open-source.

Mistral AI Models

Mistral offers models under Apache 2.0 (like Mistral 7B) for unrestricted commercial use, and some under their own Mistral AI License with usage restrictions. Always verify which license applies to your specific model variant.

Falcon (TII)

The Falcon series typically uses Apache 2.0, making it one of the most permissive options for commercial deployment without usage caps or special terms.

Key Considerations

  • Commercial restrictions: Can you monetize services built on this model?
  • Derivative works: Must you share fine-tuned versions?
  • Attribution requirements: Do you need to credit the creators?
  • Scale limitations: Are there user count thresholds?

Key Takeaway: Before deploying any open-weight model, read its license carefully—licenses vary even within the same model family, and violating terms can expose you to legal risk in production environments.