Measuring this metric is critical for the AI Futures Model, as research taste largely determines the speed of reaching artificial superintelligence once coding is fully automated. The human baseline used in the study consisted of experienced researchers, though most of those individuals had not worked at a frontier lab.
Tuesday, October 20, 2026
Story
Opus 5.5's Research Taste Hits 2.3x Score of Human Experts
AI14 days ago
pzero research found that the capacity for experimental taste in frontier AI models has doubled every three months since December 2025. The leading model, Opus 5.5, currently possesses 2.3 times the research taste of experienced human specialists, matching the scores of the best human experts on typical tasks using about 17 GPU hours of experiments instead of 40.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.