Task-horizon research

METR

How difficult a task can an agent complete, expressed in human task time?

metr.org ↗By METR (Model Evaluation & Threat Research)Reviewed 15 October 2026

Historical reference: the publisher says this page is no longer actively updated.

Key information

Focus: Software-oriented task difficulty and autonomous task completion.
Measure: Fitted task-completion horizon at a stated success level; not AI runtime.
Version / scope: Time Horizon 1.1; this page is no longer actively updated
Update pattern: The publisher explicitly stopped active updates to this page. It is retained as a methodological and historical reference.
Publisher update: 8 September 2026
Editorial verification: 15 October 2026

In plain English

What makes it different?

METR's time-horizon work relates agent success to how long a human expert takes to complete a task. This offers a way to describe task difficulty beyond a flat pass rate. The referenced page is now historical; it is not a current league table of all leading models.

Original scor.es editorial guide.
Sources and attribution are listed below.

What it measures

Software-oriented task difficulty and autonomous task completion.

Agents
Coding
Reliability

Strengths

Connects success rates to an interpretable task-difficulty measure.
Discusses uncertainty, task selection and limits of generalisation.

Limitations

Task durations, scaffolds and curve fits introduce assumptions.
The suite is concentrated in software-related work; coverage is incomplete. The publisher warns that its longest-horizon measurements are unreliable.

How the evaluation works

Researchers estimate human task durations and fit a relationship between duration and agent success. A chosen success level identifies the corresponding horizon.

This is about difficulty in a particular software-oriented task suite, rather than how long the AI runs or how many hours of any job it can replace.

Configuration matters

Specify the task suite, success threshold, scaffold and model setup. A 50% horizon and an 80% horizon answer different questions; the same number of hours is not general job automation.

Primary methods: Official analysis repository ↗Original time-horizon research ↗

Results at the publisher

Scores stay with the original source

We have not documented the permissions needed to reproduce this leaderboard here. You can still explore its methods and visit the publisher for results.

Open METR ↗

Versions & freshness

Verified 15 October 2026

The source labels Time Horizon 1.1 and records a September 8, 2026 notice that the page is no longer actively updated. This directory's verification date is separate.

Official research page and analysis repository accessible; leaderboard maintenance stopped. The publisher explicitly stopped active updates to this page. It is retained as a methodological and historical reference.

Our review date is not an evaluation date. Individual run dates are not consistently available; they remain unknown here. No automatic result collection is enabled.

Publication & reuse

Permission not documented

Public research and an analysis repository are available, but a scoped grant for republishing these result tables, retaining history and commercial reuse has not been established in this review.

Code access is not treated as permission for the underlying evaluation data or charts.

View the scope of the reuse decision

Result fields: No result fields cleared
Public result display: not enabled
Storage: not enabled
Historical copies: not enabled
Updates: not enabled
Commercial reuse: not enabled
Derived ratings: No scoped grant documented; not enabled.
Attribution: Credit the named publisher and link the primary sources. Attribution alone does not grant reproduction rights.

This decision does not clear mixed-source scientific recommendations. Publisher content, software, datasets and model licences are separate. No partnership is implied.

Primary sources & attribution

This guide is original scor.es commentary about METR (Model Evaluation & Threat Research)'s work. The links below support the identity, methods, versions and reuse review.

Official task-completion time-horizon page ↗Checked 15 October 2026
Official analysis repository ↗Checked 15 October 2026
Original time-horizon research ↗Checked 15 October 2026

Editorial edition 2026-10-15.1 · reuse review directory-rights-1. Re-check due 14 November 2026.