AgentPlays
Parsimony: an open-source benchmark for how much code AI models add
Show HN: Parsimony measures net code units AI models add on solved DeepSWE Python tasks, not just solve rate. Existing benchmarks ignore excess code that makes long-term maintenance harder. Live site compares models on a scatterplot against Artificial Analysis Intelligence Index.
How you’d use it: Compare coding agents by how much extra code they leave behind.