Benchmarks / Berkeley Function Calling Leaderboard (BFCL) V4
Reported by Berkeley Function Calling Leaderboard (BFCL) V4
Berkeley Function Calling Leaderboard (BFCL) V4
Making a call when a suitable function is available.
- Results dated
- 16 Dec 2025
- Models
- 109
- Unit
- % correct
- Licence
- Apache License 2.0
| # | Model | BFCL: relevance detection % correct, higher is better |
|---|---|---|
| 1 | Gemma 3 4BGoogle · gemma-3-4b-it |
100.0%
|
| 1 | Llama 3.3 70B InstructMeta · llama-3.3-70b-instruct |
100.0%
|
| 1 | llama-4-maverick-17b-128e-instruct-fp8Meta |
100.0%
|
| 1 | Llama 4 ScoutMeta · llama-4-scout |
100.0%
|
| 1 | nanbeige3.5-pro-thinkingNanbeige |
100.0%
|
| 1 | GPT-4.1OpenAI · gpt-4.1 |
100.0%
|
| 1 | falcon3-7b-instructTII |
100.0%
|
| 8 | Qwen3 235B A22B Instruct 2507Alibaba · qwen3-235b-a22b-2507 |
93.8%
|
| 8 | Qwen3 30B A3B Instruct 2507Alibaba · qwen3-30b-a3b-instruct-2507 |
93.8%
|
| 8 | Qwen3 32BAlibaba · qwen3-32b |
93.8%
|
| 8 | Qwen3 8BAlibaba · qwen3-8b |
93.8%
|
| 8 | Nova Pro 1.0Amazon · nova-pro-v1 |
93.8%
|
| 8 | DeepSeek V3.2 ExpDeepSeek · deepseek-v3.2-exp |
93.8%
|
| 8 | Gemma 3 12BGoogle · gemma-3-12b-it |
93.8%
|
| 8 | Llama 3.1 8B InstructMeta · llama-3.1-8b-instruct |
93.8%
|
| 8 | mistral-large-2411Mistral |
93.8%
|
| 8 | mistral-large-2411Mistral |
93.8%
|
| 8 | Mistral NemoMistral · mistral-nemo |
93.8%
|
| 8 | Mistral Small 3.2 24BMistral · mistral-small-3.2-24b-instruct |
93.8%
|
| 8 | GPT-4.1 NanoOpenAI · gpt-4.1-nano |
93.8%
|
| 8 | GPT-5 MiniOpenAI · gpt-5-mini |
93.8%
|
| 8 | GPT-5 NanoOpenAI · gpt-5-nano |
93.8%
|
| 8 | o3OpenAI |
93.8%
|
| 8 | bielik-11b-v2.3-instructSpeakleash And Ack Cyfronet Agh |
93.8%
|
| 8 | falcon3-10b-instructTII |
93.8%
|
| 8 | coalm-70bUiuc Oumi |
93.8%
|
| 27 | Qwen3 14BAlibaba · qwen3-14b |
87.5%
|
| 27 | Qwen3 235B A22B Instruct 2507Alibaba · qwen3-235b-a22b-2507 |
87.5%
|
| 27 | qwen3-4b-instruct-2507Alibaba |
87.5%
|
| 27 | qwen3-4b-instruct-2507Alibaba |
87.5%
|
| 27 | granite-20b-functioncallingIBM |
87.5%
|
| 27 | Llama 3.2 3B InstructMeta · llama-3.2-3b-instruct |
87.5%
|
| 27 | Mistral Small 3.2 24BMistral · mistral-small-3.2-24b-instruct |
87.5%
|
| 27 | GPT-4.1 MiniOpenAI · gpt-4.1-mini |
87.5%
|
| 27 | GPT-4.1OpenAI · gpt-4.1 |
87.5%
|
| 27 | xlam-2-1b-fc-rSalesforce |
87.5%
|
| 27 | xlam-2-3b-fc-rSalesforce |
87.5%
|
| 27 | xlam-2-8b-fc-rSalesforce |
87.5%
|
| 27 | coalm-8bUiuc Oumi |
87.5%
|
| 27 | grok-4-0709xAI |
87.5%
|
| 41 | qwen3-1.7bAlibaba |
81.2%
|
| 41 | Qwen3 14BAlibaba · qwen3-14b |
81.2%
|
| 41 | Qwen3 30B A3B Instruct 2507Alibaba · qwen3-30b-a3b-instruct-2507 |
81.2%
|
| 41 | Qwen3 32BAlibaba · qwen3-32b |
81.2%
|
| 41 | Nova Micro 1.0Amazon · nova-micro-v1 |
81.2%
|
| 41 | Command ACohere · command-a |
81.2%
|
| 41 | Gemma 3 27BGoogle · gemma-3-27b-it |
81.2%
|
| 41 | granite-4.0-350mIBM |
81.2%
|
| 41 | arch-agent-32bKatanemo |
81.2%
|
| 41 | Mistral NemoMistral · mistral-nemo |
81.2%
|
| 41 | GPT-4.1 MiniOpenAI · gpt-4.1-mini |
81.2%
|
| 41 | o3OpenAI |
81.2%
|
| 41 | o4 MiniOpenAI · o4-mini |
81.2%
|
| 41 | o4 MiniOpenAI · o4-mini |
81.2%
|
| 41 | rzn-tPhronetic Ai |
81.2%
|
| 41 | xlam-2-32b-fc-rSalesforce |
81.2%
|
| 41 | falcon3-3b-instructTII |
81.2%
|
| 41 | palmyra-x-004Writer |
81.2%
|
| 41 | grok-4-0709xAI |
81.2%
|
| 41 | grok-4-1-fast-non-reasoningxAI |
81.2%
|
| 41 | grok-4-1-fast-reasoningxAI |
81.2%
|
| 62 | qwen3-0.6bAlibaba |
75.0%
|
| 62 | qwen3-0.6bAlibaba |
75.0%
|
| 62 | Qwen3 8BAlibaba · qwen3-8b |
75.0%
|
| 62 | Nova 2 LiteAmazon · nova-2-lite-v1 |
75.0%
|
| 62 | Gemini 2.5 FlashGoogle · gemini-2.5-flash |
75.0%
|
| 62 | gemini-3-pro-previewGoogle |
75.0%
|
| 62 | toolace-2-8bHuawei Noah And Ustc |
75.0%
|
| 62 | granite-3.2-8b-instructIBM |
75.0%
|
| 62 | arch-agent-1.5bKatanemo |
75.0%
|
| 62 | hammer2.1-1.5bMadeagents |
75.0%
|
| 62 | Mistral Medium 3Mistral · mistral-medium-3 |
75.0%
|
| 62 | Kimi K2 0711Moonshot AI · kimi-k2 |
75.0%
|
| 62 | nanbeige4-3b-thinking-2511Nanbeige |
75.0%
|
| 62 | GPT-5.2OpenAI · gpt-5.2 |
75.0%
|
| 62 | GPT-5.2OpenAI · gpt-5.2 |
75.0%
|
| 62 | GPT-5 NanoOpenAI · gpt-5-nano |
75.0%
|
| 62 | xlam-2-70b-fc-rSalesforce |
75.0%
|
| 62 | GLM 4.6Z.ai · glm-4.6 |
75.0%
|
| 80 | Claude Opus 4.5Anthropic · claude-opus-4.5 |
68.8%
|
| 80 | Claude Sonnet 4.5Anthropic · claude-sonnet-4.5 |
68.8%
|
| 80 | bitagent-bounty-8bBittensor |
68.8%
|
| 80 | command-a-reasoningCohere |
68.8%
|
| 80 | Command R7B (12-2024)Cohere · command-r7b-12-2024 |
68.8%
|
| 80 | gemini-3-pro-previewGoogle |
68.8%
|
| 80 | granite-3.1-8b-instructIBM |
68.8%
|
| 80 | arch-agent-3bKatanemo |
68.8%
|
| 80 | hammer2.1-0.5bMadeagents |
68.8%
|
| 80 | GPT-4.1 NanoOpenAI · gpt-4.1-nano |
68.8%
|
| 80 | minicpm3-4bOpenbmb |
68.8%
|
| 91 | Claude Haiku 4.5Anthropic · claude-haiku-4.5 |
62.5%
|
| 91 | Claude Opus 4.5Anthropic · claude-opus-4.5 |
62.5%
|
| 91 | Gemini 2.5 FlashGoogle · gemini-2.5-flash |
62.5%
|
| 91 | Mistral Medium 3Mistral · mistral-medium-3 |
62.5%
|
| 91 | GPT-5 MiniOpenAI · gpt-5-mini |
62.5%
|
| 96 | hammer2.1-3bMadeagents |
56.2%
|
| 96 | minicpm3-4bOpenbmb |
56.2%
|
| 98 | Gemini 2.5 Flash LiteGoogle · gemini-2.5-flash-lite |
50.0%
|
| 98 | hammer2.1-7bMadeagents |
50.0%
|
| 98 | Phi 4Microsoft · phi-4 |
50.0%
|
| 101 | Gemini 2.5 Flash LiteGoogle · gemini-2.5-flash-lite |
43.8%
|
| 101 | Llama 3.2 1B InstructMeta · llama-3.2-1b-instruct |
43.8%
|
| 103 | Claude Sonnet 4.5Anthropic · claude-sonnet-4.5 |
37.5%
|
| 103 | DeepSeek V3.2 ExpDeepSeek · deepseek-v3.2-exp |
37.5%
|
| 103 | gemma-3-1b-itGoogle |
37.5%
|
| 106 | Claude Haiku 4.5Anthropic · claude-haiku-4.5 |
31.2%
|
| 107 | ministral-8b-2410Mistral |
0.0%
|
| 107 | llama-3.1-nemotron-ultra-253b-v1NVIDIA |
0.0%
|
| 107 | falcon3-1b-instructTII |
0.0%
|
Results as published by Berkeley Function Calling Leaderboard (BFCL) V4; we do not re-run them.
What it measures
Making a call when a suitable function is available.
What it does not measure
Not whether the call itself was right; only that one was attempted.
Berkeley Function Calling Leaderboard (BFCL) V4 by the UC Berkeley Gorilla team, Apache License 2.0.