Claude Fable 5.2 benchmarks tie Opus 5.5 on TypeScript monorepo

An author testing Anthropic's Claude Fable 5.2 against Opus 5.5 on a real TypeScript monorepo found that Fable tied or lost in benchmark scores primarily because exact-match tests penalized its more accurate bug fixes. The author also argues that Fable's higher security attack success rate in system card tests is actually an artifact of a high fallback rate to older models when requests are flagged by classifiers. Ultimately, the author recommends Opus 5.5 for routine tasks, Fable for high-stakes fixes and code reviews, and disabling auto-fallback. Open article →