According to CoinMeta, tests conducted by AI startup Xplore on their RWE-Bench tool showed that GPT-6 ranked first among 11 models with a single-pass average pass rate of 37%, Claude Opus with 5.5 achieved 29%, and GPT-6.1 SOL scored 25%. The tests covered 20 mechanical, electrical, and aerospace tasks, including engine crankshaft design, circuit board wiring, and turbine blade cooling simulation. Each model used its own corresponding agent toolchain to directly operate professional engineering software. All designs were automatically verified by the engineering software and physical solvers, and any failure to meet the stringent requirements was considered a failure. Xplore aims to transform these engineering tasks into a reinforcement learning training environment to enhance the engineering capabilities of AI.