BenchmarkFocusBench
FocusBench
A benchmark proposed by Meta for evaluating the Active Focus capability of agents.
Timeline (last 90 days)
Oct 12
Meta uses FocusBench to evaluate focus and SWE-Marathon to test sustained execution on long-horizon software projects.
Unverified50%
Oct 12
FocusBench is a process-based evaluation that measures focus by constructing multi-step tasks and counting the proportion of times the agent continues, switches, asks, confirms, or stops at the right moment.
Unverified50%
All Facts (2)
Unverified
Meta uses FocusBench to evaluate focus and SWE-Marathon to test sustained execution on long-horizon software projects.
50%UnverifiedFocusBench is a process-based evaluation that measures focus by constructing multi-step tasks and counting the proportion of times the agent continues, switches, asks, confirms, or stops at the right moment.
50%