First Impressions with the Claude 3 Opus Vision API

Anthropic's Claude 3, released on March 4, 2024, is a new series of multimodal models that reportedly surpasses competitors like GPT-4 with Vision in language and vision tasks. The Roboflow team conducted a series of tests on the Claude 3 Opus API to assess its capabilities. The model excelled in Optical Character Recognition (OCR) for reading text on images and performed well in some visual question answering tasks, such as identifying movie scenes. However, it showed limitations in tasks like object detection and currency counting and notably refused to perform OCR on text mentioning celebrities due to copyright concerns. Despite some promising results, the model struggled with certain tasks that other models have successfully completed, reflecting the challenges faced by multimodal models in general.