Berkeley Function-Calling Leaderboard (BFCL) V4

Berkeley Function-Calling Leaderboard (BFCL) V4 is run by UC Berkeley (Gorilla / Shishir Patil et al.). It has ranked 109 models, of which the catalogue holds 25, scoring from 51.4 to 77.47 on Overall Acc.

measured by UC Berkeley (Gorilla / Shishir Patil et al.) · the board itself

Place	Model	Metric	Score
1stof 109	Claude Opus 4.5	Overall Acc (fc)	77.47
2ndof 109	Claude Sonnet 4.5	Overall Acc (fc)	73.24
3rdof 109	gemini-3-pro-preview	Overall Acc (prompt)	72.51
4thof 109	GLM-4.6	Overall Acc (fc thinking)	72.38
5thof 109	Grok 4.1 Fast Reasoning	Overall Acc (fc)	69.57
6thof 109	Claude Haiku 4.5	Overall Acc (fc)	68.7
7thof 109	gemini-3-pro-preview	Overall Acc (fc)	68.14
8thof 109	o3	Overall Acc (prompt)	63.05
9thof 109	Grok 4	Overall Acc (prompt)	62.97
10thof 109	Grok 4	Overall Acc (fc)	61.38
11thof 109	Kimi K2 0711	Overall Acc	59.06
12thof 109	Grok 4.1 Fast Non-Reasoning	Overall Acc (fc)	58.29
13thof 109	Command A	Overall Acc (fc, reasoning)	57.06
14thof 109	DeepSeek V3.2 Exp	Overall Acc (prompt + thinking)	56.73
15thof 109	Gemini 2.5 Flash	Overall Acc (fc)	56.24
16thof 109	GPT-5.2	Overall Acc (fc)	55.87
17thof 109	GPT-5 mini	Overall Acc (fc)	55.46
18thof 109	xLAM-2-32b-fc-r	Overall Acc (fc)	54.66
19thof 109	DeepSeek V3.2 Exp	Overall Acc (fc)	54.12
20thof 109	GPT-4.1	Overall Acc (fc)	53.96
21stof 109	o4-mini	Overall Acc (fc)	53.24
22ndof 109	Llama-xLAM-2-70b-fc-r	Overall Acc (fc)	53.07
23rdof 109	Qwen3-235B-A22B-Instruct-2507	Overall Acc	52.15
24thof 109	GPT-5 nano	Overall Acc (fc)	51.45
25thof 109	Nanbeige4-3B-Thinking-2511	Overall Acc (fc)	51.4

Place

Model

Metric

Score

1stof 109

Claude Opus 4.5

Overall Acc (fc)

77.47

2ndof 109

Claude Sonnet 4.5

Overall Acc (fc)

73.24