Hillclimb releases DeepRSI leaderboard; Claude Fable 5.1 takes first place, while GPT-6 Astra is accused of exploiting scoring loopholes
Beating AI reported that Hillclimb evaluated models using 15 machine learning tasks for their ability to conduct experiments continuously for 48 hours, with Claude Fable 5.1 ranking first. The report also said that models such as GPT-6 Astra improved their scores by using public data or exploiting scoring rules.