Task horizon (time horizon)
The length of a software task, measured in human expert time, that an AI can complete with a given success rate (usually 50%).
METR measures the 50%-success task horizon across software tasks and finds it doubling roughly every 7 months since 2019 — and every ~3 months on SWE-bench Verified alone. It is the cleanest single measure of agent progress because it is independent of any one benchmark. A horizon of hours means multi-file features; days would mean whole projects.