Llama 3.2 11B Vision Instruct

Instruction-tuned 11B Llama 3.2 Vision model for image reasoning, visual question answering, document understanding, and captioning. NVIDIA NIM lists text plus image input, text output, and a 128K context window for the Llama 3.2 Vision collection.

Llama 3.2 11B Vision Instruct is Meta's entry-level multimodal model, released in September 2024 as part of the Llama 3.2 family. With 11 billion parameters, it was among the first openly available Meta models to accept image inputs alongside text, supporting a 128,000-token combined context window for text and image content. The model produces text-only output. NVIDIA NIM documents it as accepting text plus image input with text output within the Llama 3.2 Vision collection's shared context limit.

The instruction-tuned variant is fine-tuned for visual question answering, image captioning, document understanding, and figure interpretation in both single-turn and multi-turn conversational settings. It uses the same Llama 3 tokenizer and base architecture as the text-only Llama 3.2 models, extended with a vision encoder that projects image patches into the language model's embedding space.

Llama 3.2 11B Vision Instruct is available as open weights under Meta's Llama Community License and hosted on OpenRouter, Fireworks AI, NVIDIA NIM, AWS Bedrock, Azure AI Foundry, and Bitdeer. Teams needing stronger visual reasoning at the cost of higher compute should evaluate the Llama 3.2 90B Vision Instruct variant, which shares the same architecture and context window but has substantially more parameters.

Llama 3.2 11B Vision Instruct has a 128k-token context window.

Llama 3.2 11B Vision Instruct input tokens at $0.049/1M, output at $0.676/1M.