Netgen

A new benchmark called MecEng, introduced in an arXiv paper (2608.14615v1), evaluates large language models on mechanical engineering tasks, finding that the best open-weight model achieves an 86.0% s…