Benchmarks / Berkeley Function Calling Leaderboard (BFCL) V4
Reported by Berkeley Function Calling Leaderboard (BFCL) V4
Berkeley Function Calling Leaderboard (BFCL) V4
Storing and recalling facts across sessions with memory tools.
- Results dated
- 16 Dec 2025
- Models
- 109
- Unit
- % correct
- Licence
- Apache License 2.0
| # | Model | BFCL: memory % correct, higher is better |
|---|---|---|
| 1 | Claude Opus 4.5Anthropic · claude-opus-4.5 |
73.8%
|
| 2 | Claude Sonnet 4.5Anthropic · claude-sonnet-4.5 |
65.0%
|
| 3 | gemini-3-pro-previewGoogle |
61.7%
|
| 4 | grok-4-0709xAI |
55.9%
|
| 5 | GLM 4.6Z.ai · glm-4.6 |
55.7%
|
| 6 | gemini-3-pro-previewGoogle |
54.8%
|
| 7 | Claude Haiku 4.5Anthropic · claude-haiku-4.5 |
54.4%
|
| 8 | DeepSeek V3.2 ExpDeepSeek · deepseek-v3.2-exp |
54.2%
|
| 9 | grok-4-1-fast-reasoningxAI |
54.0%
|
| 10 | o3OpenAI |
51.8%
|
| 11 | grok-4-0709xAI |
50.5%
|
| 12 | o3OpenAI |
47.3%
|
| 13 | GPT-5.2OpenAI · gpt-5.2 |
45.8%
|
| 14 | nanbeige3.5-pro-thinkingNanbeige |
45.2%
|
| 15 | GPT-5 MiniOpenAI · gpt-5-mini |
44.3%
|
| 16 | DeepSeek V3.2 ExpDeepSeek · deepseek-v3.2-exp |
44.1%
|
| 17 | Gemini 2.5 FlashGoogle · gemini-2.5-flash |
41.3%
|
| 18 | Gemini 2.5 FlashGoogle · gemini-2.5-flash |
38.7%
|
| 19 | nanbeige4-3b-thinking-2511Nanbeige |
36.8%
|
| 20 | o4 MiniOpenAI · o4-mini |
35.3%
|
| 21 | o4 MiniOpenAI · o4-mini |
34.2%
|
| 22 | GPT-5 MiniOpenAI · gpt-5-mini |
29.2%
|
| 23 | Kimi K2 0711Moonshot AI · kimi-k2 |
29.0%
|
| 24 | command-a-reasoningCohere |
28.8%
|
| 25 | Gemma 3 12BGoogle · gemma-3-12b-it |
27.5%
|
| 25 | falcon3-10b-instructTII |
27.5%
|
| 27 | GPT-4.1 MiniOpenAI · gpt-4.1-mini |
26.9%
|
| 28 | Qwen3 32BAlibaba · qwen3-32b |
26.7%
|
| 29 | grok-4-1-fast-non-reasoningxAI |
26.2%
|
| 30 | mistral-large-2411Mistral |
24.9%
|
| 31 | Phi 4Microsoft · phi-4 |
24.7%
|
| 31 | GPT-5 NanoOpenAI · gpt-5-nano |
24.7%
|
| 33 | GPT-5 NanoOpenAI · gpt-5-nano |
24.5%
|
| 34 | GPT-4.1 MiniOpenAI · gpt-4.1-mini |
24.3%
|
| 35 | Qwen3 235B A22B Instruct 2507Alibaba · qwen3-235b-a22b-2507 |
23.9%
|
| 35 | qwen3-4b-instruct-2507Alibaba |
23.9%
|
| 35 | GPT-4.1OpenAI · gpt-4.1 |
23.9%
|
| 38 | mistral-large-2411Mistral |
23.7%
|
| 39 | Mistral Medium 3Mistral · mistral-medium-3 |
23.0%
|
| 40 | Mistral Medium 3Mistral · mistral-medium-3 |
21.7%
|
| 41 | GPT-4.1OpenAI · gpt-4.1 |
21.5%
|
| 42 | xlam-2-32b-fc-rSalesforce |
20.9%
|
| 43 | Gemini 2.5 Flash LiteGoogle · gemini-2.5-flash-lite |
20.6%
|
| 43 | falcon3-7b-instructTII |
20.6%
|
| 45 | Qwen3 14BAlibaba · qwen3-14b |
19.6%
|
| 46 | Qwen3 235B A22B Instruct 2507Alibaba · qwen3-235b-a22b-2507 |
19.4%
|
| 47 | llama-4-maverick-17b-128e-instruct-fp8Meta |
18.9%
|
| 47 | GPT-4.1 NanoOpenAI · gpt-4.1-nano |
18.9%
|
| 49 | toolace-2-8bHuawei Noah And Ustc |
18.5%
|
| 50 | Mistral Small 3.2 24BMistral · mistral-small-3.2-24b-instruct |
18.1%
|
| 51 | Qwen3 30B A3B Instruct 2507Alibaba · qwen3-30b-a3b-instruct-2507 |
17.6%
|
| 51 | qwen3-4b-instruct-2507Alibaba |
17.6%
|
| 53 | GPT-4.1 NanoOpenAI · gpt-4.1-nano |
16.8%
|
| 54 | Command ACohere · command-a |
16.6%
|
| 55 | Qwen3 32BAlibaba · qwen3-32b |
15.7%
|
| 56 | Mistral Small 3.2 24BMistral · mistral-small-3.2-24b-instruct |
15.1%
|
| 57 | Qwen3 8BAlibaba · qwen3-8b |
14.6%
|
| 57 | arch-agent-32bKatanemo |
14.6%
|
| 59 | granite-3.1-8b-instructIBM |
14.4%
|
| 59 | xlam-2-70b-fc-rSalesforce |
14.4%
|
| 61 | xlam-2-8b-fc-rSalesforce |
14.0%
|
| 62 | Gemma 3 27BGoogle · gemma-3-27b-it |
13.6%
|
| 63 | Qwen3 8BAlibaba · qwen3-8b |
13.1%
|
| 63 | palmyra-x-004Writer |
13.1%
|
| 65 | Gemini 2.5 Flash LiteGoogle · gemini-2.5-flash-lite |
12.7%
|
| 66 | granite-3.2-8b-instructIBM |
12.5%
|
| 67 | minicpm3-4bOpenbmb |
12.0%
|
| 68 | xlam-2-3b-fc-rSalesforce |
11.4%
|
| 68 | bielik-11b-v2.3-instructSpeakleash And Ack Cyfronet Agh |
11.4%
|
| 70 | Qwen3 14BAlibaba · qwen3-14b |
11.2%
|
| 71 | Llama 3.1 8B InstructMeta · llama-3.1-8b-instruct |
10.8%
|
| 72 | Mistral NemoMistral · mistral-nemo |
10.3%
|
| 73 | Qwen3 30B A3B Instruct 2507Alibaba · qwen3-30b-a3b-instruct-2507 |
9.7%
|
| 74 | minicpm3-4bOpenbmb |
9.5%
|
| 75 | qwen3-0.6bAlibaba |
8.6%
|
| 75 | Gemma 3 4BGoogle · gemma-3-4b-it |
8.6%
|
| 75 | Mistral NemoMistral · mistral-nemo |
8.6%
|
| 78 | qwen3-0.6bAlibaba |
8.4%
|
| 79 | arch-agent-1.5bKatanemo |
8.2%
|
| 79 | Llama 3.3 70B InstructMeta · llama-3.3-70b-instruct |
8.2%
|
| 79 | Llama 4 ScoutMeta · llama-4-scout |
8.2%
|
| 82 | falcon3-3b-instructTII |
7.7%
|
| 83 | arch-agent-3bKatanemo |
6.9%
|
| 83 | rzn-tPhronetic Ai |
6.9%
|
| 85 | Llama 3.2 3B InstructMeta · llama-3.2-3b-instruct |
6.2%
|
| 86 | qwen3-1.7bAlibaba |
6.0%
|
| 87 | falcon3-1b-instructTII |
5.8%
|
| 87 | coalm-70bUiuc Oumi |
5.8%
|
| 89 | Claude Sonnet 4.5Anthropic · claude-sonnet-4.5 |
5.4%
|
| 90 | Command R7B (12-2024)Cohere · command-r7b-12-2024 |
5.2%
|
| 91 | ministral-8b-2410Mistral |
4.5%
|
| 92 | GPT-5.2OpenAI · gpt-5.2 |
3.9%
|
| 92 | xlam-2-1b-fc-rSalesforce |
3.9%
|
| 94 | gemma-3-1b-itGoogle |
3.2%
|
| 94 | granite-4.0-350mIBM |
3.2%
|
| 94 | Llama 3.2 1B InstructMeta · llama-3.2-1b-instruct |
3.2%
|
| 97 | hammer2.1-3bMadeagents |
3.0%
|
| 98 | coalm-8bUiuc Oumi |
2.8%
|
| 99 | Claude Haiku 4.5Anthropic · claude-haiku-4.5 |
2.6%
|
| 100 | Nova 2 LiteAmazon · nova-2-lite-v1 |
2.4%
|
| 100 | Nova Micro 1.0Amazon · nova-micro-v1 |
2.4%
|
| 102 | Nova Pro 1.0Amazon · nova-pro-v1 |
1.9%
|
| 102 | Claude Opus 4.5Anthropic · claude-opus-4.5 |
1.9%
|
| 104 | bitagent-bounty-8bBittensor |
1.5%
|
| 105 | hammer2.1-0.5bMadeagents |
1.1%
|
| 106 | granite-20b-functioncallingIBM |
0.0%
|
| 106 | hammer2.1-1.5bMadeagents |
0.0%
|
| 106 | hammer2.1-7bMadeagents |
0.0%
|
| 106 | llama-3.1-nemotron-ultra-253b-v1NVIDIA |
0.0%
|
Results as published by Berkeley Function Calling Leaderboard (BFCL) V4; we do not re-run them.
What it measures
Storing and recalling facts across sessions with memory tools.
What it does not measure
Not long-term personal memory in a product: sessions are BFCL's scripted ones.
Berkeley Function Calling Leaderboard (BFCL) V4 by the UC Berkeley Gorilla team, Apache License 2.0.