The Visual Physics Comprehension Test (VPCT) evaluates vision models' ability to predict visually simple ball-falling scenarios. Currently, humans score 100% while the top model GPT-5 (high) only scores 66%.

This market resolves to the quarter in which the first AI model achieves at least 95% accuracy on the VPCT benchmark, as reported on the official VPCT leaderboard (https://cbrower.dev/vpct) or the author decides to replace/deprecate the benchmark.