Skip to content

METR Task Horizon Growth

An interactive chart showing the exponential growth of AI task completion capabilities, based on research from METR (Model Evaluation and Threat Research).

View Full Screen

Overview

The METR methodology measures AI capabilities by the task horizon—the duration of a human task that an AI system can complete with 50% probability of success. This metric provides a practical, real-world measure of AI capability growth.

Key Findings

The 7-Month Doubling Time

Analysis of frontier model performance reveals a consistent exponential growth pattern:

  • GPT-2 (Feb 2019): 2.4 minutes
  • GPT-4 (Mar 2023): 322 minutes (5.4 hours)
  • GPT-5 (Aug 2025): 8,239 minutes (137 hours)

This represents a 3,400× improvement over 78 months, or approximately 11 doublings.

Projection to 2030

If the 7-month doubling continues:

Date Task Horizon
Aug 2025 5.7 days
Oct 2026 23 days
Dec 2027 92 days
Jan 2030 915 days (~2.5 years)

Interactive Features

  • Scale Toggle: Switch between linear and logarithmic views
  • Projection Toggle: Show/hide the 2030 extrapolation
  • Hover: See model names and exact values

Implications for Education

This growth rate has profound implications for educational content creation:

  • Current AI can handle tasks lasting days
  • By 2027, multi-week autonomous projects become feasible
  • Content creation costs approach zero as AI handles longer creative tasks

Reference