Infant Agent: A Tool-Integrated, Logic-Driven Agent with Cost-Effective API Usage

Despite the impressive capabilities of large language models (LLMs), they currently exhibit two primary limitations, \textbf{\uppercase\expandafter{\romannumeral 1}}: They struggle to $\mathbf{\text{autonomously}}\mathbf{\text{solve}}\mathbf{\text{the}}\mathbf{\text{real}}\mathbf{\text{world}}\mathbf{\text{engineering}}\mathbf{\text{problem}}$. \textbf{\uppercase\expandafter{\romannumeral 2}}: They remain $\mathbf{\text{challenged}}\mathbf{\text{in}}\mathbf{\text{reasoning}}\mathbf{\text{through}}\mathbf{\text{complex}}\mathbf{\text{logic}}\mathbf{\text{problems}}$. To address these challenges, we developed the $\text{textsc}InfantAgent$, integrating task-aware functions, operators, a hierarchical management system, and a memory retrieval mechanism. Together, these components enable large language models to sustain extended reasoning processes and handle complex, multi-step tasks efficiently, all while significantly reducing API costs. Using the $\text{textsc}InfantAgent$, GPT-4o’s accuracy on the SWE-bench-lite dataset rises from $\mathbf{0.33}\mathbf{\%}$ to $\mathbf{30}\mathbf{\%}$, and in the AIME-2024 mathematics competition, it increases GPT-4o’s accuracy from $\mathbf{13.3}\mathbf{\%}$ to $\mathbf{37}\mathbf{\%}$.