Meta presented the first two models from the Llama 4 multimodal family: Llama 4 Scout and Llama 4 Maverick. There will be four models in the family in total. Two of them have already been launched, and the rest will appear later. According to Meta, all of them were trained on a large amount of unannotated text, images, and videos to gain extensive visual understanding capabilities.