SPU benchmark tests whether AI can predict real wet-lab outcomes in 116 tasks

liangchen5518 · x · 2026-11-16

A new benchmark, Science Prediction Universe (SPU), asks whether AI can predict real wet-lab outcomes before seeing the results. Spanning multiple scientific disciplines, it includes 100 public tasks plus 16 private tasks. Its framing: science is about predicting reality, and this benchmark directly measures that ability.

Original post →

More from Research

Research channel →