FunReason-MT Technical Report: Advanced Data Synthesis Solution for Real-world Multi-Turn Tool-use

TL;DR

FunReason-MT tackles the data bottleneck for real-world multi-turn function calling by combining an Environment-API Graph–driven data collection, reverse hard-query synthesis, and a guided CoT refinement loop. The three-phase framework generates targeted, challenging, and coherent trajectories that surpass random sampling and MAS role-playing in diversity and difficulty. Empirical results on BFCLv3 show a 4B-size model achieving state-of-the-art performance among similar models, with further gains on BFCLv4 demonstrating agentic generalization. The work provides a robust data synthesis approach and training recipe that support reliable environment-enabled learning and improved real-world tool use.

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 data synthesis, hard query construction, 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 with targeted tool, 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. Further performance improvements on BFCLv4 confirm that FunReason-MT provides a reliable and robust source for agentic learning.

Figures (3)

• Figure 1: Performance on BFCL Single-Turn, Multi-Turn and Agentic Evaluation.
• Figure 2: We summarize the core challenges in multi-turn function-calling data generation into three aspects.
• Figure 3: Three Phases of FunReason-MT.