Open-source models are granted certain exemptions, but these are limited (especially at the systemic-risk threshold). No exemption should be assumed until your form of use and distribution have been analysed.
The limits are sharp. The AI Act exempts open-source general-purpose models from some provider obligations but not from all: a model carrying systemic risk falls outside the exemption altogether, and the duties to comply with EU copyright law and to publish a summary of the training data bind you whatever the licence says. Build the model into your own product and a separate set of deployer duties arises on top of that. So the licence tells you little until your actual use and distribution are mapped against those carve-outs.
Shall we apply this matter to your situation?
Tell us your specific situation in a few sentences; we'll assess it with the right team.