AI Benchmark Tests if Models Can Identify Unique Optimal Solutions

A Kaggle benchmark evaluated whether AI models can determine if an optimization problem has a single best solution or multiple equally optimal ones. It used 87 small, exhaustively solvable problems across five types to provide exact ground truth data. Models were scored on three criteria: finding an optimal solution, correctly stating its uniqueness, and precisely counting the number of optimal solutions. In the primary test, Gemini 3.7 Flash significantly outperformed gpt-oss-120b, scoring 0.90 versus 0.55 accuracy on a fixed set of 29 instances. The study found that counting multiple optima is difficult for models, with accuracy dropping sharply as the number of optimal solutions increased.