Skip to content

Benchmark

BFCL Memory

Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages.

Models measured
49
Scores recorded
76
Sources
1
IQ weight
1.16

Score distribution

24%
Mean
20.7%
Median
17.7%
Std dev
3.9%
10th percentile
54%
90th percentile
0%
Min
0%73.8%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.16 for chat models, measured over 49 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude Opus 4.5fc· Berkeley Function-Calling Leaderboard data73.8%
2Claude Sonnet 4.5fc· Berkeley Function-Calling Leaderboard data65%
3Gemini 3 Pro PreviewPrompt· Berkeley Function-Calling Leaderboard data61.7%
4Grok 4fc· Berkeley Function-Calling Leaderboard data55.9%
5GLM 4.6FC thinking· Berkeley Function-Calling Leaderboard data55.7%
6Gemini 3 Pro Previewfc· Berkeley Function-Calling Leaderboard data54.8%
7Claude Haiku 4.5fc· Berkeley Function-Calling Leaderboard data54.4%
8DeepSeek V3.2 Expfc· Berkeley Function-Calling Leaderboard data54.2%
9Grok 4.1 Fastfc· Berkeley Function-Calling Leaderboard data54%
10OpenAI o3Prompt· Berkeley Function-Calling Leaderboard data51.8%
11Grok 4Prompt· Berkeley Function-Calling Leaderboard data50.5%
12OpenAI o3fc· Berkeley Function-Calling Leaderboard data47.3%
13GPT 5.2fc· Berkeley Function-Calling Leaderboard data45.8%
14GPT-5 minifc· Berkeley Function-Calling Leaderboard data44.3%
15DeepSeek V3.2 ExpPrompt + Thinking· Berkeley Function-Calling Leaderboard data44.1%
16Gemini 2.5 Flashfc· Berkeley Function-Calling Leaderboard data41.3%
17Gemini 2.5 FlashPrompt· Berkeley Function-Calling Leaderboard data38.7%
18o4 MiniPrompt· Berkeley Function-Calling Leaderboard data35.3%
19o4 Minifc· Berkeley Function-Calling Leaderboard data34.2%
20GPT-5 miniPrompt· Berkeley Function-Calling Leaderboard data29.3%
21Kimi K2 Instructfc· Berkeley Function-Calling Leaderboard data29%
22Command Afc· Berkeley Function-Calling Leaderboard data28.8%
23Gemma 3 12B ITPrompt· Berkeley Function-Calling Leaderboard data27.5%
24GPT-4.1 minifc· Berkeley Function-Calling Leaderboard data26.9%
25Qwen3 32Bfc· Berkeley Function-Calling Leaderboard data26.7%
26Mistral Largefc· Berkeley Function-Calling Leaderboard data25%
27GPT-5 nanofc· Berkeley Function-Calling Leaderboard data24.7%
28Phi 4Prompt· Berkeley Function-Calling Leaderboard data24.7%
29GPT-5 nanoPrompt· Berkeley Function-Calling Leaderboard data24.5%
30GPT-4.1 miniPrompt· Berkeley Function-Calling Leaderboard data24.3%
31GPT 4.1fc· Berkeley Function-Calling Leaderboard data23.9%
32Qwen3 235B A22B Instruct 2507fc· Berkeley Function-Calling Leaderboard data23.9%
33Qwen3 4B Instruct 2507Prompt· Berkeley Function-Calling Leaderboard data23.9%
34Mistral LargePrompt· Berkeley Function-Calling Leaderboard data23.7%
35Mistral Medium 3fc· Berkeley Function-Calling Leaderboard data23%
36Mistral Medium 3· Berkeley Function-Calling Leaderboard data21.7%
37GPT 4.1Prompt· Berkeley Function-Calling Leaderboard data21.5%
38Gemini 2.5 Flash-Litefc· Berkeley Function-Calling Leaderboard data20.7%
39Qwen3 14Bfc· Berkeley Function-Calling Leaderboard data19.6%
40Qwen3 235B A22B Instruct 2507Prompt· Berkeley Function-Calling Leaderboard data19.4%
41GPT-4.1 nanofc· Berkeley Function-Calling Leaderboard data18.9%
42Llama 4 Maverickfc· Berkeley Function-Calling Leaderboard data18.9%
43Mistral Smallfc· Berkeley Function-Calling Leaderboard data18.1%
44Qwen3 30B A3B Instructfc· Berkeley Function-Calling Leaderboard data17.6%
45Qwen3 4B Instruct 2507fc· Berkeley Function-Calling Leaderboard data17.6%
46GPT-4.1 nanoPrompt· Berkeley Function-Calling Leaderboard data16.8%
47Qwen3 32BPrompt· Berkeley Function-Calling Leaderboard data15.7%
48Mistral SmallPrompt· Berkeley Function-Calling Leaderboard data15.1%
49Qwen3 8Bfc· Berkeley Function-Calling Leaderboard data14.6%
50Granite 3.1 8B Instructfc· Berkeley Function-Calling Leaderboard data14.4%

Showing 1–50 of 76 results.

Data credits: Berkeley Function-Calling Leaderboard (Gorilla, UC Berkeley), Apache-2.0 (licence)