GLM-4.5V is a 106B parameter multimodal model with 12B activated parameters. Supports vision, image, video, text, and file inputs with SOTA performance across 42 visual multimodal benchmarks. Includes "Thinking Mode" for balancing speed and reasoning depth.