Claude improves existing solutions for 72 MIPLIB problems... Anthropic announces

Anthropic said its Claude model improved the registered best-known solutions on 72 of 203 open MIPLIB mixed-integer programming benchmark problems and proved optimality on 45 of them, with the results submitted to the public MIPLIB library. MIPLIB is a widely used benchmark collection for comparing optimization solvers on scheduling and resource-allocation problems, so the claim positions a general-purpose AI model against specialized mathematical optimization software. The company disclosed the figures in an official Claude social media post on the 10th and said it had submitted the solutions to MIPLIB, while noting that its own reported numbers should be distinguished from whether the public benchmark records have actually been updated. What to watch next is whether MIPLIB formally reflects the submissions in its published records, which would independently verify the gains.