P-Zero Research introduced TasteVal. It is a benchmark for experimental research taste in AI models. Research taste is the ability to pick problems, design experiments and interpret results. TasteVal measures compute efficiency on eight AI R&D tasks. Models design experiments while a fixed coding agent runs them on one H100. A run ends after 40 GPU-hours or 120 wall-clock hours. The human baseline uses the best expert attempt from 24 experts. Frontier models' experimental research taste has doubled every 3.0 months since December 2025. Opus 5.5 exceeds the human baseline with a 2.30x compute multiplier. The model costs about 1/30 of the expert baseliners' average per attempt. The paper notes this may overstate or understate the improvement rate. It does not measure a model's ability to choose problems. A naive extrapolation suggests faster progress toward superintelligence.

• doubling every 3.0 months since Dec 2025
• 95% CI 1.7-5.0 months
• Opus 5.5 compute multiplier 2.30x
• 95% CI 1.15-4.37x