JobBench - release-reported mean task reward
Professional Work
Mean rubric reward over 65 professional tasks spanning 35 occupations. The release does not establish JobBench Main's macro weighting, so this component retains the explicitly described task-mean metric.
Top models (higher is better)
| Model | Score |
|---|---|
| Muse Wonder Full | 75.3 |
| Muse Wonder | 70.5 |