The same 28 open-weight models, scored two ways: domain knowledge across 18 professional fields, and agentic ability across 20 service-desk tasks. Put side by side, they diverge: the best knowers are not the best doers. Open each full map below, or scan every model's twin profile.
Each model plotted by mean domain-capability score (x) against mean agentic score (y). The dashed line is parity. Points above it punch above their knowledge on real agent work (tool-tuned models); points below know more than they can act on.
Every model's domain fingerprint (18 cells) next to its agentic fingerprint (20 cells). Greener = stronger. Hover a cell for the axis and score.