A Toolbox, Not a Hammer -- Multi-TAG: Scaling Math Reasoning with Multi-Tool Aggregation

Augmenting large language models (LLMs) with external tools is a promising avenue for developing high-performance mathematical reasoning systems. Prior tool-augmented approaches typically finetune an LLM to select and invoke a single tool at each reasoning step and show promising results on simpler math reasoning benchmarks such as GSM8K. However, these approaches struggle with more complex math problems that require precise reasoning over multiple steps. To address this limitation, in this work, we propose Multi-T AG, a Multi-Tool AGgregation-based framework. Instead of relying on a single tool, Multi-T AG guides an LLM to concurrently invoke multiple tools at each reasoning step. It then aggregates their diverse outputs to verify and refine the reasoning process, enhancing solution robustness and accuracy. Notably, Multi-T AGis a finetuning-free, inference-only framework, making it readily applicable to any LLM backbone, including large open-weight models which are computationally expensive to finetune and proprietary frontier models which cannot be finetuned with custom recipes. We evaluate Multi-T AGon four challenging benchmarks: MA TH500, AIME, AMC, and OlympiadBench. Across both open-weight and closed-source LLM backbones, Multi-T AGconsistently and substantially outperforms state-of-the-art baselines, achieving average improvements of 6.0% to 7.5% over state-of-the-art baselines. Large Language Models (LLMs) have demonstrated remarkable capabilities across a wide variety of tasks, with reasoning emerging as a core area of research (Jiang et al., 2023; OpenAI, 2023; 2022; Y ang et al., 2024). In particular, imbuing LLMs with the ability to perform complex mathematical reasoning remains an active challenge (Ahn et al., 2024). To address this challenge, tool-augmented LLM (T ALM) frameworks such as Program-Aided Language Models (P AL) (Gao et al., 2023), Program-of-Thought (PoT) (Chen et al., 2022), Tool-Integrated Reasoning Agent (ToRA) (Gou et al., 2024), and MA THSENSEI (Das et al., 2024) equip LLMs with external tools such as Python code execution or WolframAlpha querying.