UniToolCall: Unifying Tool-Use Representation, Data, and Evaluation for LLM Agents

TL;DR

UniToolCall introduces a comprehensive framework for standardizing tool-use representation, data, and evaluation in LLM agents, significantly enhancing their tool interaction capabilities.

Contribution

It unifies the entire tool learning pipeline, curates a large tool pool, models diverse interaction patterns, and converts benchmarks into a unified evaluation format.

Findings

Achieves 93.0% single-turn Strict Precision on Hybrid-20 benchmark.

Substantially improves tool-use performance after fine-tuning Qwen3-8B.

Outperforms commercial models like GPT, Gemini, and Claude.

Abstract

Tool-use capability is a fundamental component of LLM agents, enabling them to interact with external systems through structured function calls. However, existing research exhibits inconsistent interaction representations, largely overlooks the structural distribution of tool-use trajectories, and relies on incompatible evaluation benchmarks. We present UniToolCall, a unified framework for tool learning that standardizes the entire pipeline from toolset construction and dataset generation to evaluation. The framework curates a large tool pool of 22k+ tools and constructs a hybrid training corpus of 390k+ instances by combining 10 standardized public datasets with structurally controlled synthetic trajectories. It explicitly models diverse interaction patterns, including single-hop vs. multi-hop and single-turn vs. multi-turn, while capturing both serial and parallel execution structures. To support coherent multi-turn reasoning, we further introduce an Anchor Linkage mechanism that enforces cross-turn dependencies. Furthermore, we convert 7 public benchmarks into a unified Query--Action--Observation--Answer (QAOA) representation with fine-grained evaluation at the function-call, turn, and conversation levels. Experiments show that fine-tuning Qwen3-8B on our dataset substantially improves tool-use performance. Under the distractor-heavy Hybrid-20 setting, achieves 93.0% single-turn Strict Precision, outperforming commercial models including GPT, Gemini, and Claude.