SCIENCE UNDER AUTOMATION

METR task-completion time horizons

Agents are evaluated on self-contained software, ML and cybersecurity tasks calibrated by human completion time.

What this result supports

A fitted 50% success horizon is neither an independent working-day duration nor a field takeover rate. The live page was checked without importing a frontier headline number.

Evidence record: Primary evaluation methodology; live page checked 21 October. Checked: 2026-10-21.

Read the primary source ↗

Where this evidence is used

Computer science

• Interpret a supplied specification: Repository tasks provide the brief; agents can work through it, with uneven reliability.