Anthropic reports Fable 5.2 and Mythos 5.2 match or beat recent models on alignment

Anthropic says Fable 5.2 and Mythos 5.2 performed as well as or better than its recent models on most alignment measures.

The company's X post does not name the measures or give scores. It says external evaluators, including AISecurityInst, FrontierDesign2 and METR, tested both models before release.

Andrew Curran says Anthropic's most capable model remains kept internal, citing the Fable/Mythos 5.2 System Card. His post gives no benchmark scores, release date or access details.