Benchmarks / Berkeley Function Calling Leaderboard (BFCL) V4

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

Berkeley Function Calling Leaderboard (BFCL) V4

Answering multi-step questions with a web-search tool.

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

Berkeley Function Calling Leaderboard (BFCL) V4 in brief

What it measures. Berkeley Function Calling Leaderboard (BFCL) V4 is a third-party benchmark, shown here with credit. Measured here: Answering multi-step questions with a web-search tool.

Latest results. , 81 models ranked.

Top 3 on Web search (% correct, higher is better, each model at its best setting):

  1. Claude Opus 4.5 (Anthropic): 84.5%
  2. Claude Haiku 4.5 (Anthropic): 83.5%
  3. Grok 4.1 Fast (reasoning) (xAI): 82.5%

How to cite. Berkeley Function Calling Leaderboard (BFCL) V4, via Spring Prompt, results dated 16 December 2025. https://springprompt.com/benchmarks/bfcl These results belong to Berkeley Function Calling Leaderboard (BFCL) V4, not to us. Licence: Apache License 2.0. Credit Berkeley Function Calling Leaderboard (BFCL) V4 first.

Web search: Falcon3 3B Instruct

Top 15 of 81 results · % correct, higher is better. Choose a model to highlight it.Clear highlight

  1. 1 Claude Opus 4.5Function callingAnthropic 84.5%
  2. 2 Claude Haiku 4.5Function callingAnthropic 83.5%
  3. 3 Grok 4.1 Fast (reasoning)Function callingxAI 82.5%
  4. 4 GPT-5 miniFunction callingOpenAI 82.0%
  5. 4 Grok 4 (0709)Function callingxAI 82.0%
  6. 6 Claude Sonnet 4.5Function callingAnthropic 81.0%
  7. 7 Gemini 3 Pro PreviewPromptingGoogle 80.0%
  8. 8 GLM-4.6 (reasoning on)Function callingZ.ai 77.5%
  9. 9 o3Function callingOpenAI 77.0%
  10. 10 GPT-5.2Function callingOpenAI 75.5%
  11. 10 o4-miniFunction callingOpenAI 75.5%
  12. 12 Grok 4.1 Fast (non-reasoning)Function callingxAI 75.0%
  13. 13 GPT-5 nanoFunction callingOpenAI 72.5%
  14. 14 DeepSeek-V3.2-ExpFunction callingDeepSeek 69.5%
  15. 15 GPT-4.1Function callingOpenAI 68.0%
  16. 56 Falcon3 3B InstructFunction callingTII 1.0%

Full results

BFCL: web search, % correct, higher is better
#ModelWeb search
% correct, higher is better
Price
$ per million tokens, in / out
1 Claude Opus 4.5Function callingAnthropic · best of 2 settings
84.5%
$5 / $25
2 Claude Haiku 4.5Function callingAnthropic · best of 2 settings
83.5%
$1 / $5
3 Grok 4.1 Fast (reasoning)Function callingxAI
82.5%
–
4 GPT-5 miniFunction callingOpenAI · best of 2 settings
82.0%
$0.25 / $2
4 Grok 4 (0709)Function callingxAI · best of 2 settings
82.0%
–
6 Claude Sonnet 4.5Function callingAnthropic · best of 2 settings
81.0%
$3 / $15
7 Gemini 3 Pro PreviewPromptingGoogle · best of 2 settings
80.0%
–
8 GLM-4.6 (reasoning on)Function callingZ.ai
77.5%
$0.50 / $2
9 o3Function callingOpenAI · best of 2 settings
77.0%
$2 / $8
10 GPT-5.2Function callingOpenAI · best of 2 settings
75.5%
$1.75 / $14
10 o4-miniFunction callingOpenAI · best of 2 settings
75.5%
–
12 Grok 4.1 Fast (non-reasoning)Function callingxAI
75.0%
–
13 GPT-5 nanoFunction callingOpenAI · best of 2 settings
72.5%
$0.05 / $0.40
14 DeepSeek-V3.2-ExpFunction callingDeepSeek · best of 2 settings
69.5%
$0.27 / $0.41
15 GPT-4.1Function callingOpenAI · best of 2 settings
68.0%
$2 / $8
16 Kimi K2 (0711)Function callingMoonshot AI
66.5%
$0.57 / $2.30
17 Gemini 2.5 FlashPromptingGoogle · best of 2 settings
62.0%
$0.30 / $2.50
18 GPT-4.1 miniFunction callingOpenAI · best of 2 settings
57.0%
$0.40 / $1.60
19 Command A ReasoningFunction callingCohere
55.5%
–
20 Qwen3-235B-A22B-Instruct-2507Function callingAlibaba · best of 2 settings
54.0%
$0.15 / $0.75
21 Command AFunction callingCohere
46.5%
$2.50 / $10
22 Nanbeige3.5-Pro-ThinkingFunction callingNanbeige
42.0%
–
23 Mistral Medium 3Mode not statedMistral AI · best of 2 settings
39.0%
$0.40 / $2
24 Mistral Small 3.2 24BFunction callingMistral AI · best of 2 settings
31.0%
$0.094 / $0.25
25 Llama 4 Maverick 17B 128E Instruct (FP8)Function callingMeta
28.0%
–
25 Mistral Large 2.1 (2411)Function callingMistral AI · best of 2 settings
28.0%
–
27 Command R7B (12-2024)Function callingCohere
27.0%
$0.037 / $0.15
28 Qwen3-32BPromptingAlibaba · best of 2 settings
26.0%
$0.14 / $0.40
29 xLAM-2-32b-fc-rFunction callingSalesforce
25.5%
–
30 Qwen3-30B-A3B-Instruct-2507Function callingAlibaba · best of 2 settings
22.5%
$0.09 / $0.30
31 Nanbeige4-3B-Thinking (2511)Function callingNanbeige
21.5%
–
32 Gemini 2.5 Flash-LiteFunction callingGoogle · best of 2 settings
21.0%
$0.10 / $0.40
33 xLAM-2-70b-fc-rFunction callingSalesforce
15.0%
–
34 Llama 4 ScoutFunction callingMeta
14.5%
$0.10 / $0.30
35 Qwen3-8BPromptingAlibaba · best of 2 settings
13.5%
–
36 GPT-4.1 nanoFunction callingOpenAI · best of 2 settings
11.0%
$0.10 / $0.40
37 Qwen3-14BPromptingAlibaba · best of 2 settings
10.5%
$0.10 / $0.22
38 Llama 3.3 70B InstructFunction callingMeta
10.0%
$0.71 / $0.71
39 ToolACE-2-8BFunction callingHuawei Noah's Ark Lab and USTC
8.5%
–
40 Mistral NemoFunction callingMistral AI · best of 2 settings
7.0%
$0.029 / $0.03
41 xLAM-2-8b-fc-rFunction callingSalesforce
6.5%
–
42 Nova 2 LiteFunction callingAmazon
5.0%
$0.30 / $2.50
42 Arch-Agent-32BMode not statedKatanemo
5.0%
–
44 Qwen3-4B-Instruct-2507PromptingAlibaba · best of 2 settings
4.5%
–
44 Phi-4PromptingMicrosoft
4.5%
$0.07 / $0.14
46 Gemma 3 12BPromptingGoogle
4.0%
$0.05 / $0.15
47 Llama 3.1 8B InstructPromptingMeta
3.0%
$0.05 / $0.08
48 Qwen3-1.7BFunction callingAlibaba
2.5%
–
48 Nova Pro 1.0Function callingAmazon
2.5%
$0.80 / $3.20
48 xLAM-2-3b-fc-rFunction callingSalesforce
2.5%
–
48 Palmyra X 004Function callingWriter
2.5%
–
52 MiniCPM3-4BPromptingOpenBMB · best of 2 settings
2.0%
–
53 Nova Micro 1.0Function callingAmazon
1.5%
$0.035 / $0.14
53 Bielik 11B v2.3 InstructPromptingSpeakLeash and ACK Cyfronet AGH
1.5%
–
53 Falcon3 10B InstructFunction callingTII
1.5%
–
56 Qwen3-0.6BFunction callingAlibaba · best of 2 settings
1.0%
–
56 Gemma 3 4BPromptingGoogle
1.0%
$0.04 / $0.05
56 Llama 3.2 3B InstructFunction callingMeta
1.0%
$0.05 / $0.33
56 Ministral 8B (2410)Function callingMistral AI
1.0%
–
56 Falcon3 3B InstructFunction callingTII
1.0%
–
61 Granite 3.1 8B InstructFunction callingIBM
0.5%
–
61 Granite 3.2 8B InstructFunction callingIBM
0.5%
–
61 Granite 4.0 350MFunction callingIBM
0.5%
–
61 Arch-Agent-3BMode not statedKatanemo
0.5%
–
61 Falcon3 7B InstructFunction callingTII
0.5%
–
66 BitAgent Bounty 8BMode not statedBittensor
0.0%
–
66 Gemma 3 1BPromptingGoogle
0.0%
–
66 Gemma 3 27BPromptingGoogle
0.0%
$0.08 / $0.35
66 Granite 20B FunctionCallingFunction callingIBM
0.0%
–
66 Arch-Agent-1.5BMode not statedKatanemo
0.0%
–
66 Hammer2.1-0.5BFunction callingMadeAgents
0.0%
–
66 Hammer2.1-1.5BFunction callingMadeAgents
0.0%
–
66 Hammer2.1-3BFunction callingMadeAgents
0.0%
–
66 Hammer2.1-7BFunction callingMadeAgents
0.0%
–
66 Llama 3.2 1B InstructFunction callingMeta
0.0%
$0.027 / $0.20
66 Llama 3.1 Nemotron Ultra 253B v1Function callingNVIDIA
0.0%
–
66 RZN-TPromptingPhronetic AI
0.0%
–
66 xLAM-2-1b-fc-rFunction callingSalesforce
0.0%
–
66 Falcon3 1B InstructFunction callingTII
0.0%
–
66 CoALM 70BMode not statedUIUC and Oumi
0.0%
–
66 CoALM 8BMode not statedUIUC and Oumi
0.0%
–

Swipe the table sideways for more columns.

Source: Berkeley Function Calling Leaderboard (BFCL) V4, via Spring Prompt, springprompt.com/benchmarks/bfcl. Test your own prompt on these models →

Ranks follow the score as shown, so equal numbers share a rank. Each model is shown at its best setting; show every setting. Results as published by Berkeley Function Calling Leaderboard (BFCL) V4; we do not re-run them.

Progress over time

Web search, 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%50%100% Jul 20242025Jul2026Jul Mistral NemoMistral Large 2.1 (2411)GPT-4.1o3GPT-5 miniClaude Haiku 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.

AnthropicxAIOpenAIGoogleZ.aiDeepSeek

0%50%100% Jul 20242025Jul2026Jul AnthropicxAIOpenAIGoogleZ.aiDeepSeek

What it measures

Answering multi-step questions with a web-search tool.

What it does not measure

Not general research quality: questions have short, checkable answers.

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