Function Calling and Agentic AI in 2025: What the Latest Benchmarks Tell Us About Model Performance

Function calling and agentic AI performance is evaluated using specialized benchmarks that surpass traditional metrics. The Berkeley Function Calling Leaderboard (BFCL) assesses capabilities like multi-step reasoning and tool selection, with models like GLM-4.5 and Claude 4.1 showing strong results. A more rigorous benchmark, MCPMark, stress-tests models in realistic, multi-step workflows, revealing significant performance gaps. On MCPMark, GPT-5 leads but only achieves a 52.6% success rate, indicating that even top models struggle with complex, real-world agentic tasks.