token-bench

Measure what it costs an AI coding agent to work in your codebase. An agent reads a whole file to change any part of it, so token-bench reports that per-file edit cost in Claude's own tokens and flags which files are worth splitting.