FunReason-MT Technical Report: Overcoming the Complexity Barrier in Multi-Turn Function Calling

2025-10-28

AI Summary

Existing function-calling (FC) data synthesis methods struggle to support multi-turn tool-interaction training, facing three key challenges: poor target-model adaptability, architectural isolation across tools, and strong logical dependencies across turns. This paper proposes an environment-driven framework for generating high-quality multi-turn FC data. Our approach innovatively integrates environment-API graph modeling, high-level tool query synthesis, and guided iterative chain-of-thought (CoT) generation—overcoming traditional limitations in logical coherence and tool decoupling. Specifically, we construct diverse and high-fidelity training data via graph-structured environment modeling, multi-turn dialogue synthesis, reinforcement-learning-based trajectory sampling, and CoT derivation. Evaluated on the BFCL benchmarks, a 4B-parameter model trained on our synthetic data achieves state-of-the-art performance among models of comparable scale on BFCLv3 and consistently outperforms most closed-source models on BFCLv4.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Machine Learning: Large Multimodal Models (LMMs)
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Function calling (FC) empowers large language models (LLMs) and autonomous agents to interface with external tools, a critical capability for solving complex, real-world problems. As this ability becomes increasingly central to advanced AI systems, the need for high-quality, multi-turn training data to develop and refine it cannot be overstated. Existing data synthesis methods, such as random environment sampling or multi-agent role-playing, are not powerful enough to generate high-quality data in real-world environments. Practical challenges come in three folds: targeted model training, isolation of tool architecture, and multi-turn logical dependency. To address these structural deficiencies, we present FunReason-MT, a novel data synthesis framework for real-world multi-turn tool use. FunReason-MT resolves the complexity barrier in multi-turn FC data by employing 1) Environment-API Graph Interactions to gather varied high-quality trajectories, 2) Advanced Tool-Query Synthesis to simplify hard query construction, and 3) Guided Iterative Chain for sophisticated CoT generation. Evaluations on Berkeley Function-Calling Leaderboard (BFCLv3) demonstrate the power of our framework: a 4B model built upon FunReason-MT generated data achieves state-of-the-art performance among comparable-sized models, outperforming most close-source models. Further performance improvements on BFCLv4 confirm that FunReason-MT provides a reliable and robust source for agentic learning.

Problem

• Addressing structural deficiencies in multi-turn function calling data synthesis
• Overcoming complexity barriers in real-world tool use training
• Resolving challenges in targeted model training and logical dependencies

Innovation

• Environment-API Graph Interactions for trajectory collection
• Advanced Tool-Query Synthesis simplifies query construction
• Guided Iterative Chain generates sophisticated reasoning steps