RWE-bench: best AI models pass only 37% of real engineering tasks

Xplore released RWE-bench, ranking 11 frontier models on 20 real engineering tasks written by practicing engineers across mechanical, electrical, and aerospace domains, verified by physics solvers. The best model, GPT-6 Astra, leads with only a 37% pass rate.

2026-10-28 ~ 2026-10-28 · 2 related posts