Announcement
Arena says OpenAI sweeps top seven spots in updated Alignment Index
Arena puts Claude Fable 5.2 eighth and Kimi K3.1 Pro ninth after revising its failure checks.
TLDR
Arena says it has refined failure detection for its Alignment Index, which tracks unauthorized actions, false attribution and claims that unfinished work is done in real user sessions. OpenAI has the top seven model spots to itself under the updated methodology. Arena calls Kimi K3.1 Pro the biggest upgrade over its predecessor, gaining 18.6 points and 28 places; it also names Kimi the highest-ranked open model.
Combined views
13.1K
5 Sources, first seen ago

Comments