AI TrackerGet the app
← All news
OpenAIModels

Epoch AI: no model meets its staff standard on open-ended work tasks; Fable 5.2 leads with 82%, GPT-6.2 Astra is fourth with 65%

Epoch AI tested GPT-6.2 Astra and other recent models on Epoch Automation Reports, its benchmark of open-ended Epoch work tasks, from making Epoch-style diagrams to designing basic research experiments. An Epoch employee scores each output against the quality standard expected of Epoch staff, where 100% meets it.

Epoch says every model fell short. Claude Fable 5.2 scored highest at an average of 82%, and GPT-6.2 Astra ranked fourth at 65%, even though it shares the top ECI (Epoch Capabilities Index) score of 174 with Fable 5.2. On one task, Epoch says, both Fable 5.2 and Claude Opus 5.5 matched the level of a contractor.

The scores are Epoch's own, graded by its staff. The post had 224 points on X by November 18.

Get news like this as a push.

AI Tracker alerts you when something changes for the tools you use.

Get the app