Measuring AI Ability to Complete Long Software Tasks

Despite rapid progress on AI benchmarks, the real-world meaning of benchmark performance remains unclear. To quantify the capabilities of AI systems in terms of human capabilities, we propose a new metric: 50%-task-completion time horizon. This is the time humans typically take to complete tasks that AI models can complete with 50% success rate. We first timed humans with relevant domain expertise on a combination of RE-Bench, HCAST, and 66 novel shorter tasks. On these tasks, current frontier AI models such as Claude 3.7 Sonnet have a 50% time horizon of around 50 minutes. Furthermore, frontier AI time horizon has been doubling approximately every seven months since 2019, though the trend may have accelerated in 2024. The increase in AI models' time horizons seems to be primarily driven by greater reliability and ability to adapt to mistakes, combined with better logical reasoning and tool use capabilities. We discuss the limitations of our results -- including their degree of external validity -- and the implications of increased autonomy for dangerous capabilities. If these results generalize to real-world software tasks, extrapolation of this trend predicts that within 5 years, AI systems will be capable of automating many software tasks that currently take humans a month.

The authors rated tasks for "messiness." After controlling for task length, agents performed worse on messier tasks. In this subset, however, the authors found no evidence that the improvement trend had plateaued or slowed specifically for higher-messiness tasks. On sub-hour tasks, success rates rose 40 percentage points from January 2023 to May 2025 in both the high- and low-messiness splits. [p 31 ] The authors say this does not rule out slower progress on the broader distribution of real software-engineering work.

Human completion times give benchmark results a common, intuitive scale. On the tested tasks, that scale shows a steady doubling of the 50% horizon. The authors caution that the time horizon depends on the task distribution and on the humans used as baseliners. [p 10 ] Whether the trend generalizes to real software work remains unresolved.