“Microsoft's MAI-THINKING-1 model achieved a 97% score on the AME benchmark for reasoning performance.”