We fingerprinted every image in 1,231 Codex conversations (22.83 GB). 71% is images, only 16% unique; 92% of duplicates sit in conversations used this week; forks depend on other files’ bytes. Why the honest cleanup is compression, not de-duplication.
How this was measured #
Measured on one developer’s Mac on September 2 and 3, 2026, with a byte-level pass over every rollout file in the Codex store (1,231 files, 22.83 GB, about ninety seconds), not from Mint’s own summaries. Every embedded image was hashed with SHA-256 and compared inside its conversation and against every other one; each conversation was dated by its last use. Whether a rewritten conversation still opens was checked through Codex’s own reader (the app-server’s thread/read), which is also how the broken forks were found. The Claude Code, Cursor and Claude Desktop figures come from the same kind of pass over their own directories. The scan hooks that produced them ship in Mint and are read-only.