Berkeley Function-Calling Leaderboard (BFCL)

Name: Berkeley Function-Calling Leaderboard (BFCL)
Published: 2024-02-26

Agents · 2024-02-26

Function-calling and tool-use benchmark covering simple, multiple, parallel, executable, REST/API, SQL, Java, JavaScript, and relevance-detection cases.

Top models (higher is better)

Model	Score
Claude 3.5 Sonnet (June 2024)	90.2
Llama 3.1 405B Instruct	88.5
GPT-4 0125 Preview	88.3
Llama 3.1 70B Instruct	84.8
Llama 3 70B Instruct	83.0
GPT-4o (source unspecified)	80.5
Llama 3.1 8B Instruct	76.1
QwQ-32B	66.4
o1-mini	62.8
Llama 3 8B Instruct	60.3
DeepSeek-R1	60.3
DeepSeek-R1-Distill-Qwen-32B	53.5
DeepSeek-R1-Distill-Llama-70B	49.3