METR has introduced the "expenditure horizon," a metric designed to quantify the cost-effectiveness threshold where AI agents become more expensive than hiring humans to solve specific problems. The metric assigns a dollar value that represents the point at which computational costs exceed human labor costs for task completion.
Early testing on the NanoGPT speedrun benchmark reveals disappointing results. Current AI agents remain substantially more expensive than human workers for many tasks, suggesting widespread deployment as cost-effective labor replacements remains distant. The metric calculates this by measuring the total computational expenditure required for an agent to solve a problem versus the salary cost of a human worker performing the same task.
The expenditure horizon addresses a real gap in AI capability measurement. Existing benchmarks focus on accuracy and speed but ignore economic viability. This new approach directly ties performance to real-world deployment decisions. An agent that solves a problem faster but consumes more compute resources may still fail the cost test.
However, the metric carries notable limitations. It assumes static labor costs and ignores variable factors like infrastructure overhead, training time, and human error rates. The benchmark also struggles with task complexity comparisons. Speedrun tasks may not reflect real-world economic dynamics where human workers handle exception cases and require supervision.
The timing matters. Newer model generations with improved efficiency could dramatically shift these calculations. A model requiring 50 percent less compute to reach the same performance level changes the expenditure horizon significantly. METR's framework provides a standardized way to track these shifts over time.
The real value lies in transparency. As AI companies race to scale agents, having a public metric that measures economic viability forces honest assessment of actual deployment readiness. Current results suggest the hype around AI agents replacing human workers outpaces the underlying economics. The expenditure horizon lets investors and researchers track when that changes.