Credit-Budgeted ICPC-Style Coding: When Agents Must Pay for Every Decision

TL;DR

This paper introduces USACOArena, a resource-constrained environment for evaluating autonomous coding agents, emphasizing cost-aware problem-solving and strategic trade-offs in a strict credit economy.

Contribution

It presents a novel interactive arena that models real-world resource constraints, enabling development of efficient, cost-aware coding agents and highlighting current challenges in balancing accuracy with resource limits.

Findings

Current agents fail to optimize accuracy-resource trade-offs.

Agents exhibit divergent, path-dependent behaviors under constraints.

USACOArena serves as a dynamic training ground for resource-aware architectures.

Abstract

Current evaluations of autonomous coding agents assume an unrealistic, infinite-resource environment. However, real-world software engineering is a resource-bound competition. As we scale toward large agent swarms, ignoring compute and time costs risks catastrophic budget exhaustion. To shift the focus from isolated accuracy to cost-aware problem-solving, we introduce USACOArena, an interactive ACM-ICPC-style arena driven by a strict "credit" economy. Every generated token, local test, and elapsed second depletes a fixed budget, forcing agents to make strategic trade-offs. Our comprehensive profiling reveals that frontier single agents and swarms currently fail to optimally balance accuracy with these constraints, exhibiting divergent, path-dependent behaviors. Ultimately, USACOArena provides an essential dynamic training ground for developing highly efficient, resource-aware agent architectures.