Fable 5.2 Drops False Positives by 55% but Downgrades to Opus 4.8
TL;DR
Anthropic’s release of Fable 5.2 introduces notable improvements, including a 55% reduction in false positives from its safety classifiers. However, these advancements come with significant trade-offs. In testing, Fable 5.2 downgraded a user to the older Opus 4.8 model when tasked with creating an application hardening plan, underscoring the restrictive nature of its safety measures. […] The post Fable 5.2 Drops False Positives by 55% but Downgrades to Opus 4.8 appeared first on Geeky Gadgets.
Nauti's Take
The 55 percent figure is difficult to assess without a published test method, especially because Geeky Gadgets appears to be the only source so far. Small teams should reproduce three real hardening tasks before switching and measure when Fable blocks, downgrades, or routes the request to an older model.
Video
Sources