BLINK

BLINK is a multimodal benchmark focused on core visual perception abilities that tests whether Large Language Models can truly perceive visual content rather than just see it. The benchmark evaluates fundamental visual understanding capabilities not found in other evaluations, assessing how well models can process and interpret visual information. BLINK addresses the gap between visual input processing and genuine visual comprehension in multimodal AI systems.