The Llama 3.2 90B Vision Instruct model is a large-scale multimodal AI with 90 billion parameters, designed to handle both text and image inputs effectively. This model excels in visual recognition, image reasoning, and captioning tasks, leveraging a unique architecture that integrates image encoder representations into the language model through cross-attention layers. This allows it to perform complex tasks such as visual question answering and document understanding, including interpreting charts and graphs. Despite its impressive size, recent evaluations indicate that it performed similarly to its base version across various datasets, raising questions about the effectiveness of instruction fine-tuning in this context.