The study covers twelve checkpoints from nine architecture families; nine checkpoints receive uniform budget sweeps, four core checkpoints receive matched-budget dynamic-rule comparisons, and three additional Qwen3 checkpoints support single-factor comparisons. · CiteArk
The study covers twelve checkpoints from nine architecture families; nine checkpoints receive uniform budget sweeps, four core checkpoints receive matched-budget dynamic-rule comparisons, and three additional Qwen3 checkpoints support single-factor comparisons.
来源：fixed-paper:PDF pp. 4–5, Sections 3.2–3.3 and Table 1
