Harbor Frontier Bench

Harbor Frontier Bench (HFB) evaluates and compares frontier AI agents across a range of benchmark suites — spanning coding, reasoning, math, and science. It aggregates results from 10 suites covering 1,373 individual tasks into a unified view of agent capability.

Agents Under Evaluation

Models Under Evaluation

Datasets Under Evaluation

Benchmark Categories

Tasks span 6 categories covering the breadth of what these agents are asked to do.

Methodology

Each benchmark suite reports its own metric — exact match, accuracy, or pass rate. An agent's score on a suite is the mean reward across that suite's tasks, scaled to a percentage. The radar, bar, and heatmap views compare agents across the suites they share.