Intel / article

Expenditure Horizon: Measuring Optimization Ability, with an Application to NanoGPT

Source: metr.org
Date: Published Jul 21, 2026

Why it matters

Offers a cost-based way to measure when agents beat humans at optimization work, with real figures from NanoGPT, which is useful for budgeting -driven R&D.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.

Read the source metr.org