Can I use an open-weight AI model commercially?

Not automatically. "Open weights" means you can download the weights, not that the license permits commercial use. And the license tag you see is often wrong: a model's terms inherit through its fine-tune lineage, so a model tagged permissive can really be a fine-tune of a non-commercial base two hops up. To use one safely, trace the lineage to the original base, read each license in the chain, and take the most restrictive term, not the tag.

Updated 19 June 2026

The tag is a claim, not a contract

The license field is metadata a human set, and on community fine-tunes it inherits wrong constantly. The binding terms are in the actual license files of the model and everything it's derived from.

Licenses inherit through fine-tunes

A fine-tune is a derivative of the base weights, so it carries the base's terms. A restriction does not wash out by fine-tuning, if the base is non-commercial, the fine-tune is too, whatever tag it wears.

How to clear a model before you ship

Trace to the root, take the strictest term, check the dataset license too, read for use-based clauses (non-commercial, no-compete, field-of-use), and write down the result so an auditor or acquirer can see it was cleared.

Straight answers.

Does a fine-tuned model inherit its base's license?: Usually yes, it's a derivative of the base weights, so it's bound by the base's terms plus anything the fine-tuner adds. Relabelling it doesn't make it commercial.
Why does this matter for a company?: Discovering at a fundraise or acquisition that you shipped on a non-commercial model is expensive. Clearing the license costs an hour; ripping a mislicensed model out of a shipped product costs a quarter.
Are 'open' model licenses the same as open source?: Often no. Many add acceptable-use, no-commercial, or no-compete clauses a true OSI-open license never would. Read for them.

Keep reading.