Fable 5.2 and Mythos 5.2 perform comparably to or better than our recent models on most alignment metrics.
Fable 5.2 and Mythos 5.2 perform comparably to or better than our recent models on most alignment metrics.
The model was also tested by external evaluators before release, including @AISecurityInst, @FrontierDesign2, and @METR_Evals.