BFCL-V4

BFCL (Berkeley Function Calling Leaderboard) v4 is a collection of function-calling tests: simple and parallel calls, live and irreversible actions, multi-turn conversations and memory. The model must convert user requests into exactly the right tool calls — a wrong argument or a missed restriction loses points.

BFCL-V4 measures how well a model acts as a general-purpose agent. In detail, it checks:

• converting natural language into correct function calls
• parallel and nested (multi-step) tool usage
• asking back instead of guessing on ambiguous or irreversible actions
• remembering earlier constraints across turns

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.

What is the BFCL-V4 benchmark?

BFCL (Berkeley Function Calling Leaderboard) v4 is a collection of function-calling tests: simple and parallel calls, live and irreversible actions, multi-turn conversations and memory. The model must convert user requests into exactly the right tool calls — a wrong argument or a missed restriction loses points.

What does BFCL-V4 test?

BFCL-V4 mainly measures how well a model acts as a general-purpose agent. Concretely, it checks: converting natural language into correct function calls; parallel and nested (multi-step) tool usage; asking back instead of guessing on ambiguous or irreversible actions.

How is the BFCL-V4 score calculated?

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.

Where do I get BFCL-V4 data and scores?

The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/bfcl-v4/scores) or pull it directly from the Open LLM Radar API with any HTTP client.