New SPU benchmark finds frontier AI agents fail at predicting wet-lab results

UniPatAI introduced the Science Prediction Universe (SPU) benchmark, which tasks AI agents with autonomously predicting wet-lab outcomes across 100 tasks and 4207 experiments. Frontier models performed poorly, with the best average score at just 0.486.

2026-11-14 ~ 2026-11-16 · 2 related posts

• SPU benchmark: frontier models fail at predicting 4,207 real wet-lab experiments — 量子位 · 2026-11-14
• Science Prediction Universe benchmark: best agents score just 0.486 at predicting wet-lab outcomes — liangchen5518 · 2026-11-16