Acting Less is Reasoning More! Teaching Language Model to Act Efficiently

Tool-integrated reasoning (TIR) augments large language models (LLMs) with the ability to invoke external tools during long-form reasoning, such as search engines and code interpreters, to solve tasks beyond the capabilities of internal reasoning. While reinforcement learning (RL) has shown promise in training such agents, most of existing approaches typically optimize only for final correctness without considering the efficiency or necessity of external tool use. This often leads to excessive tool calling, leading to increased computational costs and additional latency, and may also shift reliance toward external tools rather than the model’s own reasoning -- a phenomenon referred to as \textit{cognitive offloading}. To this end, we propose Optimized Tool Call-controlled Policy Optimization (OTC-PO), a simple yet effective RL-based framework that encourages models to produce accurate answers with less tool calls. Our method introduces a tool-integrated reward that jointly considers answer correctness and corresponding tool use behavior of model to reach that answer. To validate the effectiveness, we introduce the additional metric of \textit{tool productivity}, defined as the ratio between the number of correct answers and the total number of tool calls across all test cases. This metric reflects how efficiently and effectively tool usage contributes to successful task completion, with higher values indicating more productive external tool calls with the help of internal reasoning. We then instantiate this framework within both Proximal Policy Optimization (PPO) and Group Relative Preference Optimization (GRPO), resulting in OTC-PPO and OTC-GRPO. Experiments with Qwen-2.5 and Qwen-Math across multiple QA benchmarks show that our approach reduces tool calls by up to 68.3\% and improves tool productivity by up to 215.4\%, while maintaining comparable answer accuracy, especially for the larger models.