Bingguang/FunReason-MT

FunReason-MT-4B is a 4 billion parameter Large Language Model developed by Bingguang, fine-tuned from Qwen3-4B-Instruct-2507. It specializes in complex, multi-turn Function Calling (FC) and agentic tool-use tasks. The model achieves superior performance on the Berkeley Function-Calling Leaderboard (BFCLv3) Multi-Turn and Agentic Evaluation benchmarks, demonstrating the effectiveness of its novel FunReason-MT data synthesis framework for generating high-quality multi-turn FC data.