Articles about web development, AI, and technology
Parsimony is a new benchmark that measures how much code AI coding agents add when solving tasks. It counts syntax units rather than lines, subtracts deleted code, and ranks models by mean net units added on SWE-bench Verified tasks.
