Drone-Bench evaluates simple drone surveillance capabilities of frontier AI models

Drone-Bench is a new evaluation suite focused on drone surveillance scenarios. It tests the ability of frontier AI models to interpret simple aerial observations. The benchmark presents a set of controlled drone footage inputs. Models are assessed on how accurately they track objects and detect activity. Results help researchers gauge current strengths and gaps in model perception. The suite is hosted on Andon Labs at the provided URL. Developers can run the benchmark against their own models. Findings aim to guide future improvements in AI-driven drone analytics.