Timeline Model release February 5, 2026 SIU-LDG-2026-009 SI Index +4.50

Claude Opus 4.6 passes a full workday on METR's task-horizon test

METR puts the 50% time horizon of Anthropic's new model at about 12 hours, past both the 8-hour workday and the 10-hour level of the work-week milestone.

Log-scale chart of METR time horizons through February 2026, with Claude Opus 4.6 highlighted at about 12 hours, above the one-workday line — METR 50% time horizons by release date, through Claude Opus 4.6. Chart: SI USA, from METR data

OpenAI’s GPT-5.3-Codex, released the same day, was measured at 5h50m. METR has warned that its current task suite cannot reliably measure horizons above about 16 hours, which limits what the Autonomy pillar can register from here.

Index treatment. Autonomous task horizon of one workday completes, and one work-week reaches its first level.

SI Index

+4.50 28.00 → 32.50

Autonomous task horizon of one workday

Autonomy & Agency · level 0.50 → 1.00 · +3.00

METR's point estimate of 11h59m clears 8 hours. It was first reported at about 14.5 hours; METR corrected a modeling bug on March 3, 2026, and the corrected figure is used.
Autonomous task horizon of one work-week

Autonomy & Agency · level 0.00 → 0.25 · +1.50

Clears the 10-hour level. The 95% interval is very wide (about 5 to 60 hours), so the next level waits for a point estimate past 20 hours.

Points are weight × change in level, computed from the methodology.

Why it matters in the U.S.

The first U.S. model measured past a full day of expert work. In November 2025 the best measured horizon was under five hours.

Sources

Introducing Claude Opus 4.6 Anthropic
Time horizons (Time Horizon 1.1 results) METR

Anthropic

The SI USA Briefing.