Benchmarks / Berkeley Function Calling Leaderboard (BFCL) V4

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

Berkeley Function Calling Leaderboard (BFCL) V4

BFCL's own weighted average across its test categories.

Data older than 90 days: last updated 16 Dec 2025

Results dated
16 Dec 2025
Results
109 configurations of 81 models
Unit
% correct
Licence
Apache License 2.0

Overall accuracy: DeepSeek-V3.2-Exp

Top 15 of 109 results · % correct, higher is better. Choose a model to highlight it.Clear highlight

  1. 1 Claude Opus 4.5Function callingAnthropic 77.5%
  2. 2 Claude Sonnet 4.5Function callingAnthropic 73.2%
  3. 3 Gemini 3 Pro PreviewPromptingGoogle 72.5%
  4. 4 GLM-4.6 (reasoning on)Function callingZ.ai 72.4%
  5. 5 Grok 4.1 Fast (reasoning)Function callingxAI 69.6%
  6. 6 Claude Haiku 4.5Function callingAnthropic 68.7%
  7. 7 Gemini 3 Pro PreviewFunction callingGoogle 68.1%
  8. 8 o3PromptingOpenAI 63.1%
  9. 9 Grok 4 (0709)PromptingxAI 63.0%
  10. 10 Grok 4 (0709)Function callingxAI 61.4%
  11. 11 Kimi K2 (0711)Function callingMoonshot AI 59.1%
  12. 12 Grok 4.1 Fast (non-reasoning)Function callingxAI 58.3%
  13. 13 Command A ReasoningFunction callingCohere 57.1%
  14. 14 DeepSeek-V3.2-Exp (reasoning on)PromptingDeepSeek 56.7%
  15. 15 Gemini 2.5 FlashFunction callingGoogle 56.2%
  16. 19 DeepSeek-V3.2-ExpFunction callingDeepSeek 54.1%

Full results

BFCL: overall accuracy, % correct, higher is better
#ModelOverall accuracy
% correct, higher is better		Price
$ per million tokens, in / out
1 Claude Opus 4.5Function callingAnthropic
77.5%
$5 / $25
2 Claude Sonnet 4.5Function callingAnthropic
73.2%
$3 / $15
3 Gemini 3 Pro PreviewPromptingGoogle
72.5%
–
4 GLM-4.6 (reasoning on)Function callingZ.ai
72.4%
$0.50 / $2
5 Grok 4.1 Fast (reasoning)Function callingxAI
69.6%
–
6 Claude Haiku 4.5Function callingAnthropic
68.7%
$1 / $5
7 Gemini 3 Pro PreviewFunction callingGoogle
68.1%
–
8 o3PromptingOpenAI
63.1%
$2 / $8
9 Grok 4 (0709)PromptingxAI
63.0%
–
10 Grok 4 (0709)Function callingxAI
61.4%
–
11 Kimi K2 (0711)Function callingMoonshot AI
59.1%
$0.57 / $2.30
12 Grok 4.1 Fast (non-reasoning)Function callingxAI
58.3%
–
13 Command A ReasoningFunction callingCohere
57.1%
–
14 DeepSeek-V3.2-Exp (reasoning on)PromptingDeepSeek
56.7%
$0.27 / $0.41
15 Gemini 2.5 FlashFunction callingGoogle
56.2%
$0.30 / $2.50
16 GPT-5.2Function callingOpenAI
55.9%
$1.75 / $14
17 GPT-5 miniFunction callingOpenAI
55.5%
$0.25 / $2
18 xLAM-2-32b-fc-rFunction callingSalesforce
54.7%
–
19 DeepSeek-V3.2-ExpFunction callingDeepSeek
54.1%
$0.27 / $0.41
20 GPT-4.1Function callingOpenAI
54.0%
$2 / $8
21 o4-miniFunction callingOpenAI
53.2%
$1.10 / $4.40
22 xLAM-2-70b-fc-rFunction callingSalesforce
53.1%
–
23 Qwen3-235B-A22B-Instruct-2507PromptingAlibaba
52.2%
$0.15 / $0.75
24 GPT-5 nanoFunction callingOpenAI
51.5%
$0.05 / $0.40
25 Nanbeige4-3B-Thinking (2511)Function callingNanbeige
51.4%
–
26 Gemini 2.5 FlashPromptingGoogle
50.9%
$0.30 / $2.50
27 GPT-4.1 miniFunction callingOpenAI
50.5%
$0.40 / $1.60
28 o4-miniPromptingOpenAI
50.3%
$1.10 / $4.40
29 Qwen3-32BFunction callingAlibaba
48.7%
$0.14 / $0.40
30 o3Function callingOpenAI
48.6%
$2 / $8
31 Qwen3-235B-A22B-Instruct-2507Function callingAlibaba
48.0%
$0.15 / $0.75
32 Nanbeige3.5-Pro-ThinkingFunction callingNanbeige
47.7%
–
33 Qwen3-32BPromptingAlibaba
46.8%
$0.14 / $0.40
34 xLAM-2-8b-fc-rFunction callingSalesforce
46.7%
–
35 Command AFunction callingCohere
46.5%
$2.50 / $10
36 BitAgent Bounty 8BMode not statedBittensor
46.2%
–
37 Arch-Agent-32BMode not statedKatanemo
45.4%
–
38 GPT-5.2PromptingOpenAI
45.3%
$1.75 / $14
39 Qwen3-8BFunction callingAlibaba
42.6%
$0.12 / $0.46
40 ToolACE-2-8BFunction callingHuawei Noah's Ark Lab and USTC
42.4%
–
41 Qwen3-30B-A3B-Instruct-2507Function callingAlibaba
41.4%
$0.09 / $0.30
42 xLAM-2-3b-fc-rFunction callingSalesforce
41.2%
–
43 Qwen3-14BFunction callingAlibaba
41.0%
$0.12 / $0.24
44 Qwen3-8BPromptingAlibaba
40.4%
$0.12 / $0.46
45 GPT-4.1PromptingOpenAI
39.4%
$2 / $8
46 Mistral Large 2.1 (2411)Function callingMistral AI
38.4%
–
47 Qwen3-14BPromptingAlibaba
37.8%
$0.12 / $0.24
48 Mistral Medium 3Mode not statedMistral AI
37.7%
$0.40 / $2
49 Mistral Medium 3Function callingMistral AI
37.6%
$0.40 / $2
50 Llama 4 Maverick 17B 128E Instruct (FP8)Function callingMeta
37.3%
–
51 Mistral Small 3.2 24BFunction callingMistral AI
37.2%
$0.094 / $0.25
52 Gemini 2.5 Flash-LiteFunction callingGoogle
36.9%
$0.10 / $0.40
53 Qwen3-30B-A3B-Instruct-2507PromptingAlibaba
36.7%
$0.09 / $0.30
54 Qwen3-4B-Instruct-2507Function callingAlibaba
35.7%
–
55 Qwen3-4B-Instruct-2507PromptingAlibaba
35.5%
–
56 Arch-Agent-3BMode not statedKatanemo
35.4%
–
57 Claude Opus 4.5PromptingAnthropic
33.5%
$5 / $25
58 GPT-4.1 nanoFunction callingOpenAI
33.1%
$0.10 / $0.40
59 Mistral Small 3.2 24BPromptingMistral AI
32.4%
$0.094 / $0.25
60 Arch-Agent-1.5BMode not statedKatanemo
32.1%
–
60 Command R7B (12-2024)Function callingCohere
32.1%
$0.037 / $0.15
62 Llama 3.3 70B InstructFunction callingMeta
31.9%
$0.59 / $0.79
63 Mistral Large 2.1 (2411)PromptingMistral AI
31.8%
–
64 Hammer2.1-7BFunction callingMadeAgents
31.7%
–
65 xLAM-2-1b-fc-rFunction callingSalesforce
30.4%
–
65 Gemma 3 12BPromptingGoogle
30.4%
$0.05 / $0.15
67 GPT-4.1 miniPromptingOpenAI
29.7%
$0.40 / $1.60
67 Hammer2.1-3BFunction callingMadeAgents
29.7%
–
69 Gemma 3 27BPromptingGoogle
29.5%
$0.12 / $0.20
70 Phi-4PromptingMicrosoft
28.8%
$0.07 / $0.14
71 Qwen3-1.7BFunction callingAlibaba
28.4%
–
72 Llama 4 ScoutFunction callingMeta
28.1%
$0.18 / $0.59
73 Gemini 2.5 Flash-LitePromptingGoogle
28.0%
$0.10 / $0.40
73 CoALM 70BMode not statedUIUC and Oumi
28.0%
–
75 Hammer2.1-1.5BFunction callingMadeAgents
27.9%
–
75 Palmyra X 004Function callingWriter
27.9%
–
77 GPT-5 miniPromptingOpenAI
27.8%
$0.25 / $2
78 Mistral NemoFunction callingMistral AI
27.6%
$0.023 / $0.03
78 GPT-5 nanoPromptingOpenAI
27.6%
$0.05 / $0.40
80 Nova 2 LiteFunction callingAmazon
27.1%
$0.30 / $2.50
80 Granite 3.1 8B InstructFunction callingIBM
27.1%
–
82 Falcon3 10B InstructFunction callingTII
27.0%
–
83 Granite 3.2 8B InstructFunction callingIBM
26.9%
–
84 CoALM 8BMode not statedUIUC and Oumi
26.8%
–
85 Llama 3.1 8B InstructPromptingMeta
25.8%
$0.05 / $0.08
86 MiniCPM3-4BFunction callingOpenBMB
25.6%
–
87 Claude Haiku 4.5PromptingAnthropic
25.3%
$1 / $5
88 Nova Pro 1.0Function callingAmazon
25.0%
$0.80 / $3.20
89 Claude Sonnet 4.5PromptingAnthropic
24.9%
$3 / $15
89 GPT-4.1 nanoPromptingOpenAI
24.9%
$0.10 / $0.40
91 Falcon3 7B InstructFunction callingTII
24.0%
–
92 Qwen3-0.6BFunction callingAlibaba
23.9%
–
93 Granite 20B FunctionCallingFunction callingIBM
23.2%
–
94 Qwen3-0.6BPromptingAlibaba
22.4%
–
95 Nova Micro 1.0Function callingAmazon
22.3%
$0.035 / $0.14
95 RZN-TPromptingPhronetic AI
22.3%
–
97 MiniCPM3-4BPromptingOpenBMB
22.1%
–
98 Llama 3.2 3B InstructFunction callingMeta
22.0%
$0.05 / $0.33
99 Bielik 11B v2.3 InstructPromptingSpeakLeash and ACK Cyfronet AGH
21.9%
–
100 Hammer2.1-0.5BFunction callingMadeAgents
21.2%
–
101 Gemma 3 4BPromptingGoogle
19.6%
$0.05 / $0.10
102 Mistral NemoPromptingMistral AI
19.3%
$0.023 / $0.03
103 Granite 4.0 350MFunction callingIBM
19.0%
–
104 Falcon3 3B InstructFunction callingTII
16.3%
–
105 Ministral 8B (2410)Function callingMistral AI
11.1%
–
105 Falcon3 1B InstructFunction callingTII
11.1%
–
107 Llama 3.2 1B InstructFunction callingMeta
10.8%
$0.027 / $0.20
108 Llama 3.1 Nemotron Ultra 253B v1Function callingNVIDIA
10.0%
–
109 Gemma 3 1BPromptingGoogle
7.2%
–

Swipe the table sideways for more columns.

Ranks follow the score as shown, so equal numbers share a rank. Results as published by Berkeley Function Calling Leaderboard (BFCL) V4; we do not re-run them.

Progress over time

Overall accuracy, plotted at each model's release date, higher is better. Each model counts once, at its best setting. Scores are today's, not launch day's; 38 models without a known release date are left off. Hover or tap a mark for its model.

Best score so far

Every model at its release date. The line moves each time a new release beats every earlier one, and each new best is named where there is room.

0%20%40%60%80%100% Jul 20242025Jul2026Jul Mistral NemoMistral Large 2.1 (2411)GPT-4.1o3Claude Sonnet 4.5Claude Opus 4.5

Best score by lab

Each lab's best so far, stepping whenever one of its releases beats its own earlier models. The 6 labs with the best scores today; 7 others not drawn.

AnthropicGoogleZ.aixAIOpenAIMoonshot AI

20%40%60%80% Jul 20242025Jul2026Jul AnthropicGoogleZ.aixAIOpenAIMoonshot AI

What it measures

BFCL's own weighted average across its test categories.

What it does not measure

Not a neutral average: the weighting is BFCL's. Not reliability on your own tools: BFCL's functions and queries are a fixed test set, and a correct call is judged by its form, not by what it achieved.

Source

Berkeley Function Calling Leaderboard (BFCL) V4 · Methodology

Berkeley Function Calling Leaderboard (BFCL) V4 by the UC Berkeley Gorilla team, Apache License 2.0.