Economics of AI · Current professional-work frontier

The Falling Cost of Useful Work

Each point compares professional-task performance on GDPval-AA v2.1 with the estimated cost of completing one benchmark task. Better models move up; cheaper models move left. The line joins the configurations that are not beaten on both dimensions.

Captured benchmark snapshots

Scores may be recalibrated between captures. Movement is indicative, not a fixed measure of absolute intelligence.

Y-axis range

Capability versus cost per task

Logarithmic cost axis · whiskers show 95% confidence intervals where published

GDPval-AA score versus benchmark cost per taskA scatter plot of current AI model configurations. Higher is better and further left is cheaper. A line connects Pareto-efficient configurations.