METR's 14h 50% Horizon Impacts The Economy More Than ASI Timelines
Claude Opus 4.6 reaches 50% coding automation in 14.5 hours, potentially reshaping economic forecasts.
The AI research organization METR released new data indicating Anthropic's Claude Opus 4.6 model has reached a 50% 'time-horizon' of approximately 14.5 hours for software engineering tasks. This metric represents the time an AI agent needs to successfully complete half of a standardized suite of coding challenges. While METR cautions the measurement is extremely noisy due to test saturation, the trend points to rapid capability growth. Forecaster Peter Wildeford extrapolates this could lead to 2-3.5 workweek horizons by end of 2026, with 'significant implications for the economy.' The core debate among AI timeline analysts (like 'AI 2027' researchers) is whether this progress in partial task automation matters more for near-term economic transformation than longer-term Artificial Superintelligence (ASI) arrival dates. A key crux is the difference between a 50% horizon (partial automation) and an 80% horizon (near-full automation), with some arguing the latter is required to bootstrap 'automated coders' that could radically accelerate research and development.
- Claude Opus 4.6 shows a 50% time-horizon of 14.5 hours on METR's software task suite, the highest point estimate yet, though with a wide confidence interval (6 to 98 hours).
- Forecasters predict this trend could lead to AI handling 2-3.5 workweeks of coding work autonomously by end of 2026, directly impacting economic productivity.
- The debate highlights a split between focusing on near-term economic impacts from partial task automation versus long-term transformative effects from full automation (80%+ horizons) needed for ASI.
Why It Matters
Accelerating AI coding autonomy could reshape software development economics years before any theoretical ASI, forcing businesses to adapt sooner.