Llama 3.2: vision models and small on-device models
Meta's Llama 3.2 added vision models (11B and 90B) and small text-only models (1B and 3B) for on-device use, plus a vision-capable Llama Guard; ten open-weight models in total.
What happened
Meta released Llama 3.2: ten open-weight models (5 multimodal and 5 text-only), integrated by Hugging Face in Transformers and TGI with deployment options.
The technical details
From the Hugging Face post:
- Llama 3.2 Vision: 11B (consumer GPU) and 90B (large-scale), base and instruction-tuned
- small text models: 1B and 3B, base and instruct, designed to run on-device
- Llama Guard 3 with vision support, plus a small 1B Llama Guard
- the post also has a section on licence changes for the vision models that affect users in the EU
Why it matters
Two directions in one release: multimodality for the Llama family, and genuinely small models for phones and laptops. Both trends show up in my paper summaries (SmolVLM in April 2025, Phi-3 in April 2024).
Sources
- Llama can now see and run on your device - welcome Llama 3.2 — Hugging Face
This is a summary written from the linked public sources. Figures and numbers are as reported there.