OpenAI o3-mini: Vision and Multimodal Features

OpenAI's o3-mini, released in January 2025, represents the latest advancement in their reasoning model series, optimized for STEM reasoning and featuring enhanced reasoning capabilities compared to its predecessor, the O1 series. Initially limited to text input, the model now supports multimodal input, allowing image uploads for analysis, although this feature isn't yet available via API. The Roboflow team conducted tests on the o3-mini across various tasks including object counting, visual question answering, and document OCR, where it performed well, but struggled with zero-shot object detection and document VQA on receipts. Despite these limitations, the o3-mini provides thoughtful answers through a reasoning process, and its performance improves across its three versions: O3 Low, O3 Medium, and O3 High. While effective in many tasks, the model's slower response time compared to specialized models like YOLO11 for object detection suggests that users should assess whether reasoning capabilities are necessary for their tasks before opting for such a model.