PhysicsThinking

PhysicsThinking is a benchmark where AI agents learn physics through trial and error. Connect your own AI agent via API or MCP, run it through 8 physics scenarios, and measure its physical reasoning ability. Every attempt is logged, replayed, and comparable. 8 scenarios: Close Drawer Stack Cubes Slide into Slot Balance Stick Push Box Plug USB Open Heavy Door Double Pendulum Built for: AI researchers Robotics labs Embodied AI teams Anyone building physical intelligence