Benchmarks

JobBench

Agentsunit % · normalized over [0, 75]

JobBench — 65 professional tasks across 35 white-collar occupations, each asking the agent to produce a practical work product scored against task-specific rubrics. Mean rubric score.

#ModelSourceScoreNormalized

Score distribution

9 tracked results across the normalization window

075

Score vs. parameters

Open-weights models, log-x params