Beth Barnes clarifies METR didn't evaluate alignment of Fable/Mythos 5.2

RyanGreenblatt · x · 2026-11-20

Beth Barnes clarifies (retweeted by Ryan Greenblatt) that METR did not evaluate alignment or communication clarity of Claude Fable 5.2 and Mythos 5.2, despite proximity in Anthropic's announcement suggesting otherwise. An eval screenshot notes Mythos 5.2 produced lengthy, unclear reports on Sunlight and, like Opus 5.5, neglected rigorous comparisons between its preferred solution and alternatives — though this weakness may matter less when the reader is another agent.

Original post →

More from Models

Models channel →