BFCL-V3

BFCL v3 is the third version of the Berkeley Function Calling Leaderboard. The focus is on stateful, multi-turn function calling in different categories (single, multiple, parallel calls) — the score is how precisely the model calls the right tools with the right parameters.

BFCL-V3 measures how well a model acts as a general-purpose agent. In detail, it checks:

• exact function-call arguments
• multi-turn state tracking
• correct selection among many similar tools
• recovery when a tool result is unexpected

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-V3 benchmark?

BFCL v3 is the third version of the Berkeley Function Calling Leaderboard. The focus is on stateful, multi-turn function calling in different categories (single, multiple, parallel calls) — the score is how precisely the model calls the right tools with the right parameters.

What does BFCL-V3 test?

BFCL-V3 mainly measures how well a model acts as a general-purpose agent. Concretely, it checks: exact function-call arguments; multi-turn state tracking; correct selection among many similar tools.

How is the BFCL-V3 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-V3 data and scores?

The official source is https://gorilla.cs.berkeley.edu/leaderboard.html. 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-v3/scores) or pull it directly from the Open LLM Radar API with any HTTP client.