Claude Haiku 5.5 fixes bugs nearly as well as Opus 5.5 at 95% lower cost

A TokenSave study on 25 real bugs from 13 open-source projects found Claude Haiku 5.5 produced 52 clean fixes out of 75 runs versus 54 for Opus 5.5, at roughly 95% lower cost ($1.41 vs $28.53). Opus broke fewer neighboring tests on hard bugs, and no model caught cases the bug report omitted.

• Haiku 5.5: 52 clean fixes of 75 runs; Opus 5.5: 54
• Cost for 75 runs: $1.41 on Haiku vs $28.53 on Opus
• On hard bugs Opus broke other tests in 5 runs, Haiku in 10
• Haiku vs Opus difference not significant (p = 0.86)