METR Adds Judgment Tasks to AI Time Horizon Chart

An updated report from the AI evaluation nonprofit METR expands its task suite and reports AI capability growth on well-defined tasks roughly doubled.

Updated on Nov. 20, 2026 in Artificial Intelligence

METR Adds Judgment Tasks to AI Time Horizon Chart

Live Poll

Do you think AI will soon handle most of your job's longer tasks?

METR, an AI evaluation nonprofit, published an updated time horizon report that adds tasks leaning harder on judgment. The update reports AI capability growth on well-defined tasks roughly doubled compared with March 2024 to October 2025.

Why it matters

METR's time horizon chart is one of the most widely shared graphics in the AI debate, so changes to its methodology shape how capability growth is read. The update also signals that performance on tasks without clear specifications is expected to be considerably lower than on well-defined tasks.

Time Horizon 1.1 expands the task set from 170 to 228 tasks, and the number of tasks taking humans eight or more hours rose substantially. METR states that results above 16 hours remain unreliable, leaving the upper end of the measurement unresolved.

The players

METR

METR is an AI evaluation nonprofit whose time horizon chart is widely shared in debates over AI progress.

Anthropic

Anthropic is an AI company whose Claude Mythos model was evaluated in METR's updated report.

The details

METR gives models a large set of tasks that human testers have already completed, and each task receives a difficulty score based on how long those testers took. The updated report adds tasks that lean harder on judgment, and its evaluation of Anthropic's Claude Mythos reported an 80% success time horizon of about 3 hours and a 50% success time horizon of approximately 17 hours.

Timeline

  1. January 29, 2026: METR publishes its Time Horizon 1.1 release, expanding the task set to 228.

  2. March 2024 to October 2025: The baseline period used to compare the rate of AI capability growth.

  3. September 8, 2026: METR leaderboard figures are updated across models.

The Tech Race

The update extends METR's time horizon series, the chart that has become a reference point in AI capability debates, by revising its task suite. It places current model results against a longer baseline running from March 2024 to October 2025.

Readers following AI product claims should note that METR says results above 16 hours remain unreliable and that real-world performance on tasks without clear specifications is expected to be considerably lower. Figures from a time horizon chart measure evaluated task suites rather than everyday usefulness.

The takeaway

A single benchmark chart can shift how the public reads AI progress, so the task mix behind any time horizon figure matters as much as the headline number. Check whether a reported capability figure reflects well-defined tasks or judgment-heavy ones.

What happens next

METR's leaderboard figures are scheduled to be updated across models on September 8, 2026.

Further reading

For broader context on how AI systems are measured, see Artificial Intelligence coverage.

Source note: This article includes information reported by Crypto Briefing.

Live Poll

Do you think AI will soon handle most of your job's longer tasks?