---
title: "METR 50% Task-Completion Horizon"
slug: the-horizon
summary: "METR's estimated 50 percent threshold: 17.4 hours of human-expert task duration"
category: "AI & Work"
unit: hours
frequency: as_published
direction: not_applicable
latest_value: 17.41
latest_period: "2026-04"
trend: rising
last_updated: "2026-09-25T00:00:00Z"
source: "METR (Model Evaluation & Threat Research)"
measurement_basis: "METR's 50 percent time horizon is the estimated human-expert task duration at which the evaluated agent's fitted success probability is 50 percent; it is not elapsed agent runtime. The current task suite primarily covers self-contained, well-specified software engineering, machine learning, and cybersecurity tasks with clear success criteria, and METR says measurements above 16 hours are unreliable."
source_url: https://metr.org/time-horizons/
canonical_html: https://americandefault.org/indicators/the-horizon/
citation_apa: "Kilburn, R. (2026). METR 50% Task-Completion Horizon [Data set]. American Default Research. https://americandefault.org/indicators/the-horizon/"
citation_chicago: "Kilburn, Ross. 2026. \"METR 50% Task-Completion Horizon.\" American Default Research. https://americandefault.org/indicators/the-horizon/."
citation_mla: "Kilburn, Ross. \"METR 50% Task-Completion Horizon.\" American Default Research, 2026, americandefault.org/indicators/the-horizon/."
---

# METR 50% Task-Completion Horizon

Also tracked as The Horizon.

> METR's estimated 50 percent threshold: 17.4 hours of human-expert task duration

## Current Reading

As of April 2026, METR 50% Task-Completion Horizon is **17.4 hours** (rising).

— benchmark tasks are primarily in software engineering, machine learning, and cybersecurity; measurements above 16 hours are unreliable

## What Changed

- METR's 50% time horizon was 17.4 hours of human-expert task time for a model released in April 2026, up from 2 hours for a model released in April 2025.
- METR says values above 16 hours (960 minutes) are unreliable on its current tasks.

## Recent Values

| Period | Value |
|--------|-------|
| 2026-04 | 17.4 hours |
| 2026-02 | 12 hours |
| 2025-12 | 5.9 hours |
| 2025-11 | 4.9 hours |
| 2025-08 | 3.4 hours |
| 2025-05 | 1.7 hours |
| 2025-04 | 2 hours |
| 2025-02 | 1 hour |
| 2024-12 | 0.7 hours |
| 2024-10 | 0.3 hours |
| 2024-09 | 0.3 hours |
| 2024-06 | 0.2 hours |

## Historical Context

- Peak since 2019: **17.4 hours** in April 2026
- Low since 2019: **0 hours** in February 2019
- Post-pandemic average (2020-01 – 2024-12): **0.2 hours**

## Methodology

This indicator carries one observation per source release rather than a fixed calendar cadence and is interpreted as measuring something other than household condition, so neither end of the range reads as household distress. Source: METR (Model Evaluation & Threat Research).
Measurement basis: METR's 50 percent time horizon is the estimated human-expert task duration at which the evaluated agent's fitted success probability is 50 percent; it is not elapsed agent runtime. The current task suite primarily covers self-contained, well-specified software engineering, machine learning, and cybersecurity tasks with clear success criteria, and METR says measurements above 16 hours are unreliable.
Automation: collected automatically on the daily refresh pipeline, with no hand entry.

## Source

METR (Model Evaluation & Threat Research) — [https://metr.org/time-horizons/](https://metr.org/time-horizons/)

Data retrieved: 2026-09-25T00:00:00Z

## Related Indicators

- [the-adoption-curve](https://americandefault.org/indicators/the-adoption-curve/)
- [the-ai-cut](https://americandefault.org/indicators/the-ai-cut/)

## Machine-Readable Endpoints

- HTML page: https://americandefault.org/indicators/the-horizon/
- JSON: https://americandefault.org/api/indicators/the-horizon.json
- CSV: https://americandefault.org/api/downloads/the-horizon.csv
- BibTeX: https://americandefault.org/citations/the-horizon.bib
- RIS: https://americandefault.org/citations/the-horizon.ris

## Citation

**APA 7th**

    Kilburn, R. (2026). METR 50% Task-Completion Horizon [Data set]. American Default Research. https://americandefault.org/indicators/the-horizon/

**Chicago (author-date)**

    Kilburn, Ross. 2026. "METR 50% Task-Completion Horizon." American Default Research. https://americandefault.org/indicators/the-horizon/.

**MLA 9th**

    Kilburn, Ross. "METR 50% Task-Completion Horizon." American Default Research, 2026, americandefault.org/indicators/the-horizon/.

Use **American Default Research** as the institutional name in source lists, bibliographies, and press citations. **American Default** is the brand and URL form. **American Distress Index (ADI)** is the product name for the national composite score.