Llama 3.2 Vision and on-device models released on Hugging Face Hub — EU licensing restriction noted

AI Impact Summary

Llama 3.2 expands the lineup with on-device text models (1B, 3B) and vision-capable variants (11B Vision and 90B Vision), plus Llama Guard 3 with optional vision safety features, all designed for offline deployment and on-device inference. Vision models support a 128k context and multilingual prompts, with tile sizes of 448 for 11B and 560 for 90B, enabling on-device visual reasoning, document QA, and image-text tasks while preserving text-model performance. EU licensing restriction blocks license rights to multimodal models for entities domiciled in the EU, creating deployment constraints even as non-EU customers can use the multimodal capabilities via the Hugging Face Hub or end-user products.