Xplore AI launches RWE-Bench for real-world engineering design
Xplore AI released RWE-Bench, a benchmark with 20 real-world engineering problems spanning mechanical, aerospace, and electronics design. Automated physical-solver checks determine whether each submitted design meets every requirement, and the reported leaderboard puts GPT-6 Astra first at 37%, followed by Claude Opus 5.5 at 29% and GPT-6.1 Sol at 25%. The benchmark was created by specialist engineers who also produced reference solutions using the same tools.
RWE-Bench provides an initial, solver-checked measure of whether AI models can perform consequential CAD and simulation work rather than only solve coding or mathematics problems.
原创 ASI启示录 2026-10-31 20:32 北京
新智元报道
AI 做数学题、写代码,已经能赢过所有人类。
可同一批模型拿到工程师每天用的 CAD 和仿真软件,每道题最多给 24 小时,去设计曲轴、给内存布线、算涡轮叶片散热,最好的 GPT-6 Astra 也只有 37% 能过。
其余模型在这个评测中,更是几乎全员挂零。
#Xplore AI#RWE-Bench#GPT-6 Astra#Claude Opus 5.5#GPT-6.1 Sol#Gemini 4 Argon#Claude Fable 5.1