Open Minis running Qwen3.8-27B just blew my mind. The model does not have vision capabilities, so it decided to OCR the images, and fulfill the request. Doesn’t seem amazing, but I was 100% expecting it to tell me that it has no vision capabilities 🤯