The largest gains from tools go mostly to models that score lowest without them. The ten models that score below 31% without tools average ~42 tool calls per task, while the seven above average ~23.
MiMo-V2.6-Pro shows the largest gain of any model: it scores 5% without tools and 43% with them, making ~49 tool calls per task. Gemini 3.8 Flash (high) makes ~68 tool calls per task and moves from 16% to 41%.
Astra and Sol make the fewest tool calls of any model, ~8 per task, and gain 13 and 16 points, respectively.