Model merging is a cost- and compute-efficient way to create models that combine the components and capabilities of existing foundational models.