Claude Opus 4.6: 14h 30m — METR confirmed Feb 21, 2026Doubling time: ~7 months overall · ~4 months since 20238-hour workday threshold: PASSED February 2026Source: METR ArXiv 2503.14499 · TH1.1 January 2026
INTERACTIVE · METR TIME HORIZON DATA · ARXIV 2503.14499
The Most Important Chart in Labor Economics
AI task complexity — measured by how long the tasks take a human expert — has doubled every 7 months since 2019. This is not a prediction. It is six years of measured data. Hover any point for details. Toggle log/linear scale. Show or hide the projection.
SCALESHOWCONFIDENCE
HUMAN TASK COMPLETION TIME (50% SUCCESS)
—
—
—
WHAT THIS MEANS
—
CONFIRMED DATA POINT
CURRENT TOP (OPUS 4.6)
EXPONENTIAL TREND (~7MO DOUBLING)
PROJECTION (2026–2028)
8H WORKDAY THRESHOLD
CURRENT TOP
14.5h
Claude Opus 4.6 · Feb 2026
OVERALL DOUBLING
7mo
2019–2024 trend · ArXiv 2503.14499
RECENT DOUBLING
4mo
123 days · 2023–present · TH1.1
8H THRESHOLD
✓
Passed February 2026 · ahead of projection
⚠ WHAT THIS CHART MEASURES — AND WHAT IT DOESN'T: The y-axis shows how long the tasks take a human expert to complete — not how long the AI spends. A "14-hour time horizon" means the AI reliably completes tasks of the complexity that take humans 14 hours. METR tests primarily software engineering, ML research, and cybersecurity tasks. This is not a direct measure of labor substitution hours. Translation to workforce displacement requires adjustment for organizational adoption lag (V9), conversion lag (V3), and task-type coverage outside software. The displacement implication is real. The translation is not 1:1. — METR researcher Sydney Von Arx, MIT Technology Review, Feb 2026.
Primary source: METR ArXiv 2503.14499 (Thomas Kwa, Ben West, Joel Becker et al., March 19, 2025) · Updated: Time Horizon 1.1 (January 2026) · Live data: metr.org/time-horizons · Wikipedia confirmed Feb 21, 2026: Claude Opus 4.6 at 14h 30m.