03 / 05
Chapter Three · The Autonomy Curve
AI can complete longer tasks.
Exponentially longer.
The measured frontier moved from a five-minute task in 2023 to roughly twelve hours in early 2026.
METR's 50% time horizon is the human-expert duration of a task that an AI agent is predicted to complete successfully half the time. It is not the amount of time an AI runs unattended, and it does not represent every kind of work. The benchmark is concentrated in software, machine learning and cybersecurity tasks.
METR Task Horizon · Task duration at 50% reliability (log scale)
The task-duration frontier.
2019 Q1
—
GPT-4 · Mar 2023
5 min
Claude 3.7 · Feb 2025
59 min
Claude Opus 4.5 · Dec 2025
4h 49m
Public frontier · Feb-Mar 2026
~12 hr
§3.1The next infrastructure cycle.
Longer task horizons are being pushed by better models, stronger agent scaffolds and more efficient infrastructure. NVIDIA says Vera Rubin is ramping into full production, with partner systems expected in the second half of 2026.
“Vera Rubin delivers 10x agent throughput at scale compared with Grace Blackwell.”
Primary sources: METR, “Measuring AI Ability to Complete Long Tasks” (Kwa et al., March 2025); the Time Horizon 1.1 dashboard, updated May 8 2026; and METR's Frontier Risk Report (May 19 2026). Hardware context: NVIDIA Vera Rubin production update (May 31 2026).