Preprint
Aug 2026
$R^3$-Bench: LLMs Struggle with Resource-Rational Reasoning under Shared Budgets
$R^3$-Bench is introduced, which evaluates six-problem suites under shared budgets across mathematics, competitive programming, and abstract reasoning in tool-free and agentic settings, and exposes a persistent gap between demonstrated competence and shared-budget realization.
Peisong Wang, Zhiwei Ma, Bo-Wen Liu et al.
· 0 citations