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.
- Results dated
- 16 Dec 2025
- Models
- 109
- Unit
- % correct
- Licence
- Apache License 2.0
| # | Model | BFCL: web search % correct, higher is better |
|---|---|---|
| 1 | Claude Opus 4.5Anthropic · claude-opus-4.5 |
84.5%
|
| 2 | Claude Haiku 4.5Anthropic · claude-haiku-4.5 |
83.5%
|
| 3 | grok-4-1-fast-reasoningxAI |
82.5%
|
| 4 | GPT-5 MiniOpenAI · gpt-5-mini |
82.0%
|
| 4 | grok-4-0709xAI |
82.0%
|
| 6 | Claude Sonnet 4.5Anthropic · claude-sonnet-4.5 |
81.0%
|
| 7 | gemini-3-pro-previewGoogle |
80.0%
|
| 8 | GLM 4.6Z.ai · glm-4.6 |
77.5%
|
| 9 | o3OpenAI |
77.0%
|
| 10 | GPT-5.2OpenAI · gpt-5.2 |
75.5%
|
| 10 | o4 MiniOpenAI · o4-mini |
75.5%
|
| 12 | grok-4-1-fast-non-reasoningxAI |
75.0%
|
| 13 | grok-4-0709xAI |
74.0%
|
| 14 | GPT-5 NanoOpenAI · gpt-5-nano |
72.5%
|
| 15 | o4 MiniOpenAI · o4-mini |
71.5%
|
| 16 | DeepSeek V3.2 ExpDeepSeek · deepseek-v3.2-exp |
69.5%
|
| 17 | gemini-3-pro-previewGoogle |
68.5%
|
| 18 | GPT-4.1OpenAI · gpt-4.1 |
68.0%
|
| 19 | Kimi K2 0711Moonshot AI · kimi-k2 |
66.5%
|
| 20 | Gemini 2.5 FlashGoogle · gemini-2.5-flash |
62.0%
|
| 21 | Gemini 2.5 FlashGoogle · gemini-2.5-flash |
59.0%
|
| 22 | DeepSeek V3.2 ExpDeepSeek · deepseek-v3.2-exp |
58.0%
|
| 23 | GPT-4.1 MiniOpenAI · gpt-4.1-mini |
57.0%
|
| 24 | command-a-reasoningCohere |
55.5%
|
| 25 | Qwen3 235B A22B Instruct 2507Alibaba · qwen3-235b-a22b-2507 |
54.0%
|
| 26 | Qwen3 235B A22B Instruct 2507Alibaba · qwen3-235b-a22b-2507 |
50.5%
|
| 26 | o3OpenAI |
50.5%
|
| 28 | Command ACohere · command-a |
46.5%
|
| 29 | nanbeige3.5-pro-thinkingNanbeige |
42.0%
|
| 30 | GPT-5.2OpenAI · gpt-5.2 |
40.5%
|
| 31 | Mistral Medium 3Mistral · mistral-medium-3 |
39.0%
|
| 32 | Mistral Medium 3Mistral · mistral-medium-3 |
35.0%
|
| 32 | GPT-4.1OpenAI · gpt-4.1 |
35.0%
|
| 34 | Mistral Small 3.2 24BMistral · mistral-small-3.2-24b-instruct |
31.0%
|
| 35 | llama-4-maverick-17b-128e-instruct-fp8Meta |
28.0%
|
| 35 | mistral-large-2411Mistral |
28.0%
|
| 37 | Command R7B (12-2024)Cohere · command-r7b-12-2024 |
27.0%
|
| 38 | Qwen3 32BAlibaba · qwen3-32b |
26.0%
|
| 39 | xlam-2-32b-fc-rSalesforce |
25.5%
|
| 40 | Qwen3 30B A3B Instruct 2507Alibaba · qwen3-30b-a3b-instruct-2507 |
22.5%
|
| 41 | Qwen3 32BAlibaba · qwen3-32b |
21.5%
|
| 41 | nanbeige4-3b-thinking-2511Nanbeige |
21.5%
|
| 43 | Gemini 2.5 Flash LiteGoogle · gemini-2.5-flash-lite |
21.0%
|
| 44 | mistral-large-2411Mistral |
20.0%
|
| 45 | Claude Haiku 4.5Anthropic · claude-haiku-4.5 |
19.5%
|
| 46 | Qwen3 30B A3B Instruct 2507Alibaba · qwen3-30b-a3b-instruct-2507 |
17.5%
|
| 47 | Claude Sonnet 4.5Anthropic · claude-sonnet-4.5 |
16.0%
|
| 48 | xlam-2-70b-fc-rSalesforce |
15.0%
|
| 49 | Llama 4 ScoutMeta · llama-4-scout |
14.5%
|
| 50 | Qwen3 8BAlibaba · qwen3-8b |
13.5%
|
| 50 | GPT-5 NanoOpenAI · gpt-5-nano |
13.5%
|
| 52 | Claude Opus 4.5Anthropic · claude-opus-4.5 |
13.0%
|
| 53 | Qwen3 8BAlibaba · qwen3-8b |
12.0%
|
| 54 | GPT-4.1 NanoOpenAI · gpt-4.1-nano |
11.0%
|
| 55 | Qwen3 14BAlibaba · qwen3-14b |
10.5%
|
| 56 | Qwen3 14BAlibaba · qwen3-14b |
10.0%
|
| 56 | Llama 3.3 70B InstructMeta · llama-3.3-70b-instruct |
10.0%
|
| 58 | toolace-2-8bHuawei Noah And Ustc |
8.5%
|
| 58 | GPT-5 MiniOpenAI · gpt-5-mini |
8.5%
|
| 60 | Mistral Small 3.2 24BMistral · mistral-small-3.2-24b-instruct |
7.5%
|
| 61 | Mistral NemoMistral · mistral-nemo |
7.0%
|
| 62 | xlam-2-8b-fc-rSalesforce |
6.5%
|
| 63 | Nova 2 LiteAmazon · nova-2-lite-v1 |
5.0%
|
| 63 | arch-agent-32bKatanemo |
5.0%
|
| 65 | qwen3-4b-instruct-2507Alibaba |
4.5%
|
| 65 | Phi 4Microsoft · phi-4 |
4.5%
|
| 67 | Gemma 3 12BGoogle · gemma-3-12b-it |
4.0%
|
| 67 | GPT-4.1 MiniOpenAI · gpt-4.1-mini |
4.0%
|
| 69 | qwen3-4b-instruct-2507Alibaba |
3.0%
|
| 69 | Llama 3.1 8B InstructMeta · llama-3.1-8b-instruct |
3.0%
|
| 71 | qwen3-1.7bAlibaba |
2.5%
|
| 71 | Nova Pro 1.0Amazon · nova-pro-v1 |
2.5%
|
| 71 | Mistral NemoMistral · mistral-nemo |
2.5%
|
| 71 | xlam-2-3b-fc-rSalesforce |
2.5%
|
| 71 | palmyra-x-004Writer |
2.5%
|
| 76 | minicpm3-4bOpenbmb |
2.0%
|
| 77 | Nova Micro 1.0Amazon · nova-micro-v1 |
1.5%
|
| 77 | GPT-4.1 NanoOpenAI · gpt-4.1-nano |
1.5%
|
| 77 | bielik-11b-v2.3-instructSpeakleash And Ack Cyfronet Agh |
1.5%
|
| 77 | falcon3-10b-instructTII |
1.5%
|
| 81 | qwen3-0.6bAlibaba |
1.0%
|
| 81 | Gemma 3 4BGoogle · gemma-3-4b-it |
1.0%
|
| 81 | Llama 3.2 3B InstructMeta · llama-3.2-3b-instruct |
1.0%
|
| 81 | ministral-8b-2410Mistral |
1.0%
|
| 81 | falcon3-3b-instructTII |
1.0%
|
| 86 | qwen3-0.6bAlibaba |
0.5%
|
| 86 | granite-3.1-8b-instructIBM |
0.5%
|
| 86 | granite-3.2-8b-instructIBM |
0.5%
|
| 86 | granite-4.0-350mIBM |
0.5%
|
| 86 | arch-agent-3bKatanemo |
0.5%
|
| 86 | falcon3-7b-instructTII |
0.5%
|
| 92 | bitagent-bounty-8bBittensor |
0.0%
|
| 92 | Gemini 2.5 Flash LiteGoogle · gemini-2.5-flash-lite |
0.0%
|
| 92 | gemma-3-1b-itGoogle |
0.0%
|
| 92 | Gemma 3 27BGoogle · gemma-3-27b-it |
0.0%
|
| 92 | granite-20b-functioncallingIBM |
0.0%
|
| 92 | arch-agent-1.5bKatanemo |
0.0%
|
| 92 | hammer2.1-0.5bMadeagents |
0.0%
|
| 92 | hammer2.1-1.5bMadeagents |
0.0%
|
| 92 | hammer2.1-3bMadeagents |
0.0%
|
| 92 | hammer2.1-7bMadeagents |
0.0%
|
| 92 | Llama 3.2 1B InstructMeta · llama-3.2-1b-instruct |
0.0%
|
| 92 | llama-3.1-nemotron-ultra-253b-v1NVIDIA |
0.0%
|
| 92 | minicpm3-4bOpenbmb |
0.0%
|
| 92 | rzn-tPhronetic Ai |
0.0%
|
| 92 | xlam-2-1b-fc-rSalesforce |
0.0%
|
| 92 | falcon3-1b-instructTII |
0.0%
|
| 92 | coalm-70bUiuc Oumi |
0.0%
|
| 92 | coalm-8bUiuc Oumi |
0.0%
|
Results as published by Berkeley Function Calling Leaderboard (BFCL) V4; we do not re-run them.
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.