Papers produced by GPT-6 Astra and Claude Fable 5.1 reportedly rejected by experts

The testers say they gave GPT-6 Astra and Claude Fable 5.1 open-ended AI research questions from unpublished papers and a $10,000 budget each. The authors of those papers graded what the models produced, and the testers say their experts rejected both papers. Compared with their first tests of Claude Opus 4.8 in July, the testers saw meaningful improvements in experimental design and resource management, but persistent limitations in creative problem solving, judgment, calibration and communication.