Anthropic reported that its Claude AI model surpassed existing top solutions on 72 complex benchmarks in the Mixed Integer Programming Library (MIPLIB), alongside delivering 45 formal optimality proofs.