2026-10-10 · America/Los_Angeles · 社区动态 · #13

I made an open-source benchmark which measures the complexity of code added (overengineering/tech debt). Looking for feedback and contributors.

Existing benchmarks measure the solve rate and token output/cost, but don''t punish excess code, which makes the codebase harder to work with in the long term. I wanted to measure this as it is especially useful for long term projects. Method: It takes the patches that benchmarks already publish (DeepSWE, SWE-bench Verified, SWE-bench Live) and counts "coding units": syntax elements like statements, calls, names and literals. I didn't measure lines of code so cramming code isn't rewarded. It also accounts for deleted code as it measures net units. GPT-6 and Claude 5.5 have no public patches…

热度 44.7 / 100；排名与评分保留该期记录。