PulseAugur
LIVE 13:06:54
research · [1 source] ·
0
research

AI's autonomous task completion time horizon doubles every 7 months

A new analysis from METR indicates that AI models' ability to autonomously complete tasks, measured by their "time horizon," is doubling approximately every 4 to 7 months across various domains. While software and reasoning tasks show horizons of 50-200+ minutes, visual computer use tasks have significantly shorter horizons but are improving at similar rates. Even in domains where AI currently excels for only seconds or minutes, exponential growth suggests potential for horizons of days or weeks within the next five years. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

RANK_REASON The cluster is based on an academic paper analyzing AI capabilities and trends.

Read on METR (Model Evaluation & Threat Research) →

AI's autonomous task completion time horizon doubles every 7 months

COVERAGE [1]

  1. METR (Model Evaluation & Threat Research) TIER_1 ·

    How Does Time Horizon Vary Across Domains?

    <figure class="breakout-wider"> <img alt="Chart of AI time horizons increasing in many domains" src="https://metr.org/assets/images/time-horizon-domains/time-horizons-increasing.png" /> <figcaption>Each line represents the trend in time horizon for one benchmark <a href="https://…