Meta releases Llama 3.2 — new multimodal models and on-device 1B/3B models

AI Impact Summary

Meta has released Llama 3.2, a new iteration of the Llama collection, bringing multimodal capabilities and smaller on-device models. The release includes 10 open-weight models, offering both multimodal and text-only options, and introduces a new Llama Guard model with vision support for input and output safety classification. The availability of 1B and 3B text models expands on-device use cases, competing with larger models in performance.