COURSE 25L1100% FREE
Verified 2026-08-14

Reading model licenses

How to read **licenses and terms** for model weights, code, and data: what you may use, modify, redistribute, and commercialize. This chapter is **literacy, not legal advice**—ask counsel for product decisions. <!-- IMAGE: checklist: lic...

What it is

How to read licenses and terms for model weights, code, and data: what you may use, modify, redistribute, and commercialize. This chapter is literacy, not legal advice—ask counsel for product decisions.

<!-- IMAGE: checklist: license file → AUP → attribution → commercial thresholds -->

HIGH PRIORITYDIAGRAM
◷ IN PRODUCTION

Visual Spec & Architecture Diagram

License layers cake: Dataset license ≠ Code license ≠ Model weight license ≠ API ToS. Crossed wires showing conflicts. Title: 'License layer cake'.

Educational Focus: Most important vendor-literacy visual.

Why it matters

“Open” is not one thing. Llama 4’s Community License (effective April 5, 2025) grants broad use/redistribution rights with attribution, Acceptable Use Policy adherence, and an additional commercial threshold for very large MAU products. Mistral documents that many open models use Apache 2.0, while some use a modified MIT with a revenue threshold for commercial licensing. Wrong assumptions create procurement pain.

How it works (plain)

  1. Find LICENSE / model card license field (Hub metadata).
  2. Find Acceptable Use / prohibited use.
  3. Check: commercial use, attribution, share-alike, training competing models, high-risk bans, territory, user-count or revenue thresholds.
  4. Note the effective date; re-verify before launch.
  5. Separate weight license from training-data licenses and from API ToS if hosted.

Everyday example

A recipe marked “share alike” vs “personal use only”—different kitchen rules even if both PDFs are free to download.

Try it

Open one model card’s license section. Write three allowed and three restricted actions in your own words, then paste the URL and date you checked.

Myths

⚠️ Myth: Weights on GitHub means public domain.
✓ Reality: Licenses still bind.
⚠️ Myth: Research-only is fine “until we scale.”
✓ Reality: Product use needs a compliant path first.
⚠️ Myth: Apache on the repo covers the weights.
✓ Reality: Code and weights can differ—read both.

Sources