Skip to content

Comment on Top AI models fail at >96% of tasksparent

Comments

They didn't test Opus at all, only Sonnet.

One of the tasks was "Build an interactive dashboard for exploring data from the World Happiness Report." -- I can't imagine how Opus4.5 could've failed that.

Check the link to the study. It has been updated for Opus 4.5.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.