搜索Ctrl K
  • 榜单数据10-22 15:50
  • 论文数据10-22 14:22
  • 周报10-19 16:07
收录成绩109109 个模型或系统
当前第一77.5Claude-Opus-4-5-20251101 (FC)
指标Overall Accuracy各评测子类准确率的未加权平均(%)
数据时效自动同步同步于 10-22 15:32 · 1 个数据源

排名表

显示 109 / 109
单轮多轮Agentic
模型机构开源来源备注
1Claude-Opus-4-5-20251101 (FC)Anthropic闭源77.588.679.868.484.573.8官方榜单2026-10-22许可证:Proprietary
2Claude-Sonnet-4-5-20250929 (FC)Anthropic闭源73.288.781.161.481.065.0官方榜单2026-10-22许可证:Proprietary
3Gemini-3-Pro-Preview (Prompt)Google闭源72.590.783.160.880.061.7官方榜单2026-10-22许可证:Proprietary
4GLM-4.6 (FC thinking)Zhipu AI开源72.487.680.968.077.555.7官方榜单2026-10-22许可证:MIT
5Grok-4-1-fast-reasoning (FC)xAI闭源69.688.378.558.982.554.0官方榜单2026-10-22许可证:Proprietary
6Claude-Haiku-4-5-20251001 (FC)Anthropic闭源68.786.578.753.683.554.4官方榜单2026-10-22许可证:Proprietary
7Gemini-3-Pro-Preview (FC)Google闭源68.185.881.763.168.554.8官方榜单2026-10-22许可证:Proprietary
8o3-2025-04-16 (Prompt)OpenAI闭源63.081.973.262.350.551.8官方榜单2026-10-22许可证:Proprietary
9Grok-4-0709 (Prompt)xAI闭源63.082.872.547.074.050.5官方榜单2026-10-22许可证:Proprietary
10Grok-4-0709 (FC)xAI闭源61.485.475.633.982.055.9官方榜单2026-10-22许可证:Proprietary
11Moonshotai-Kimi-K2-Instruct (FC)MoonshotAI开源59.181.678.750.666.529.0官方榜单2026-10-22许可证:modified-mit
12Grok-4-1-fast-non-reasoning (FC)xAI闭源58.388.177.946.875.026.2官方榜单2026-10-22许可证:Proprietary
13Command A Reasoning (FC)Cohere开源57.186.378.650.155.528.8官方榜单2026-10-22许可证:CC-BY-NC 4.0 License (w/ Acceptable Use Addendum)
14DeepSeek-V3.2-Exp (Prompt + Thinking)DeepSeek开源56.785.576.044.958.044.1官方榜单2026-10-22许可证:MIT
15Gemini-2.5-Flash (FC)Google闭源56.285.074.436.359.041.3官方榜单2026-10-22许可证:Proprietary
16GPT-5.2-2025-12-11 (FC)OpenAI闭源55.981.870.428.175.545.8官方榜单2026-10-22许可证:Proprietary
17GPT-5-mini-2025-08-07 (FC)OpenAI闭源55.569.858.627.582.044.3官方榜单2026-10-22许可证:Proprietary
18xLAM-2-32b-fc-r (FC)Salesforce开源54.789.675.569.525.520.9官方榜单2026-10-22许可证:cc-by-nc-4.0
19DeepSeek-V3.2-Exp (FC)DeepSeek开源54.134.953.737.469.554.2官方榜单2026-10-22许可证:MIT
20GPT-4.1-2025-04-14 (FC)OpenAI闭源54.082.870.038.968.023.9官方榜单2026-10-22许可证:Proprietary
21o4-mini-2025-04-16 (FC)OpenAI闭源53.237.766.141.875.534.2官方榜单2026-10-22许可证:Proprietary
22xLAM-2-70b-fc-r (FC)Salesforce开源53.188.472.277.415.014.4官方榜单2026-10-22许可证:cc-by-nc-4.0
23Qwen3-235B-A22B-Instruct-2507 (Prompt)Qwen开源52.190.378.744.650.519.4官方榜单2026-10-22许可证:apache-2.0
24GPT-5-nano-2025-08-07 (FC)OpenAI闭源51.568.059.434.572.524.7官方榜单2026-10-22许可证:Proprietary
25Nanbeige4-3B-Thinking-2511 (FC)Nanbeige开源51.481.679.451.121.536.8官方榜单2026-10-22许可证:apache-2.0
26Gemini-2.5-Flash (Prompt)Google闭源50.988.178.216.862.038.7官方榜单2026-10-22许可证:Proprietary
27GPT-4.1-mini-2025-04-14 (FC)OpenAI闭源50.583.868.834.157.026.9官方榜单2026-10-22许可证:Proprietary
28o4-mini-2025-04-16 (Prompt)OpenAI闭源50.381.370.816.671.535.3官方榜单2026-10-22许可证:Proprietary
29Qwen3-32B (FC)Qwen开源48.788.882.047.921.526.7官方榜单2026-10-22许可证:apache-2.0
30o3-2025-04-16 (FC)OpenAI闭源48.640.466.214.877.047.3官方榜单2026-10-22许可证:Proprietary
31Qwen3-235B-A22B-Instruct-2507 (FC)Qwen开源48.037.468.945.454.023.9官方榜单2026-10-22许可证:apache-2.0
32Nanbeige3.5-Pro-Thinking (FC)Nanbeige开源47.738.470.040.042.045.2官方榜单2026-10-22许可证:apache-2.0
33Qwen3-32B (Prompt)Qwen开源46.890.382.043.326.015.7官方榜单2026-10-22许可证:apache-2.0
34xLAM-2-8b-fc-r (FC)Salesforce开源46.784.668.070.06.514.0官方榜单2026-10-22许可证:cc-by-nc-4.0
35Command A (FC)Cohere开源46.587.678.529.546.516.6官方榜单2026-10-22许可证:CC-BY-NC 4.0 License (w/ Acceptable Use Addendum)
36BitAgent-Bounty-8BBittensor开源46.281.693.162.40.01.5官方榜单2026-10-22许可证:Apache-2.0
37Arch-Agent-32Bkatanemo开源45.488.980.754.35.014.6官方榜单2026-10-22许可证:katanemo-research
38GPT-5.2-2025-12-11 (Prompt)OpenAI闭源45.378.367.143.840.53.9官方榜单2026-10-22许可证:Proprietary
39Qwen3-8B (FC)Qwen开源42.687.680.541.812.014.6官方榜单2026-10-22许可证:apache-2.0
40ToolACE-2-8B (FC)Huawei Noah & USTC开源42.487.177.438.48.518.5官方榜单2026-10-22许可证:Apache-2.0
41Qwen3-30B-A3B-Instruct-2507 (FC)Qwen开源41.485.877.930.022.517.6官方榜单2026-10-22许可证:apache-2.0
42xLAM-2-3b-fc-r (FC)Salesforce开源41.283.062.958.42.511.4官方榜单2026-10-22许可证:cc-by-nc-4.0
43Qwen3-14B (FC)Qwen开源41.084.980.034.810.019.6官方榜单2026-10-22许可证:apache-2.0
44Qwen3-8B (Prompt)Qwen开源40.488.680.133.413.513.1官方榜单2026-10-22许可证:apache-2.0
45GPT-4.1-2025-04-14 (Prompt)OpenAI闭源39.488.778.99.835.021.5官方榜单2026-10-22许可证:Proprietary
46mistral-large-2411 (FC)Mistral AI闭源38.484.781.914.128.024.9官方榜单2026-10-22许可证:Proprietary
47Qwen3-14B (Prompt)Qwen开源37.889.579.326.110.511.2官方榜单2026-10-22许可证:apache-2.0
48Mistral-Medium-2505Mistral AI闭源37.785.366.09.939.021.7官方榜单2026-10-22许可证:Proprietary
49Mistral-Medium-2505 (FC)Mistral AI闭源37.667.468.010.835.023.0官方榜单2026-10-22许可证:Proprietary
50Llama-4-Maverick-17B-128E-Instruct-FP8 (FC)Meta开源37.388.773.720.328.018.9官方榜单2026-10-22许可证:Meta Llama 4 Community
51Mistral-small-2506 (FC)Mistral AI闭源37.173.677.311.531.018.1官方榜单2026-10-22许可证:Proprietary
52Gemini-2.5-Flash-Lite (FC)Google闭源36.986.665.813.521.020.6官方榜单2026-10-22许可证:Proprietary
53Qwen3-30B-A3B-Instruct-2507 (Prompt)Qwen开源36.788.978.423.517.59.7官方榜单2026-10-22许可证:apache-2.0
54Qwen3-4B-Instruct-2507 (FC)Qwen开源35.787.976.422.13.017.6官方榜单2026-10-22许可证:apache-2.0
55Qwen3-4B-Instruct-2507 (Prompt)Qwen开源35.586.474.720.54.523.9官方榜单2026-10-22许可证:apache-2.0
56Arch-Agent-3Bkatanemo开源35.486.772.934.90.56.9官方榜单2026-10-22许可证:katanemo-research
57Claude-Opus-4-5-20251101 (Prompt)Anthropic闭源33.589.776.016.113.01.9官方榜单2026-10-22许可证:Proprietary
58GPT-4.1-nano-2025-04-14 (FC)OpenAI闭源33.073.060.823.611.018.9官方榜单2026-10-22许可证:Proprietary
59Mistral-Small-2506 (Prompt)Mistral AI闭源32.489.779.014.87.515.1官方榜单2026-10-22许可证:Proprietary
60Arch-Agent-1.5Bkatanemo开源32.182.767.726.60.08.2官方榜单2026-10-22许可证:katanemo-research
61Command R7B (FC)Cohere开源32.181.069.18.327.05.2官方榜单2026-10-22许可证:cc-by-nc-4.0
62Llama-3.3-70B-Instruct (FC)Meta开源31.988.076.621.510.08.2官方榜单2026-10-22许可证:Meta Llama 3 Community
63mistral-large-2411 (Prompt)Mistral AI闭源31.883.068.113.820.023.7官方榜单2026-10-22许可证:Proprietary
64Hammer2.1-7b (FC)MadeAgents开源31.785.569.523.90.00.0官方榜单2026-10-22许可证:cc-by-nc-4.0
65xLAM-2-1b-fc-r (FC)Salesforce开源30.469.055.136.00.03.9官方榜单2026-10-22许可证:cc-by-nc-4.0
66Gemma-3-12b-it (Prompt)Google开源30.479.474.25.84.027.5官方榜单2026-10-22许可证:gemma-terms-of-use
67GPT-4.1-mini-2025-04-14 (Prompt)OpenAI闭源29.784.674.82.54.024.3官方榜单2026-10-22许可证:Proprietary
68Hammer2.1-3b (FC)MadeAgents开源29.785.070.516.50.03.0官方榜单2026-10-22许可证:qwen-research
69Gemma-3-27b-it (Prompt)Google开源29.587.274.510.80.013.6官方榜单2026-10-22许可证:gemma-terms-of-use
70Phi-4 (Prompt)Microsoft开源28.869.660.73.94.524.7官方榜单2026-10-22许可证:MIT
71Qwen3-1.7B (FC)Qwen开源28.482.974.611.02.56.0官方榜单2026-10-22许可证:apache-2.0
72Llama-4-Scout-17B-16E-Instruct (FC)Meta开源28.189.474.79.014.58.2官方榜单2026-10-22许可证:Meta Llama 4 Community
73Gemini-2.5-Flash-Lite (Prompt)Google闭源28.083.954.97.60.012.7官方榜单2026-10-22许可证:Proprietary
74CoALM-70BUIUC + Oumi开源28.083.467.310.60.05.8官方榜单2026-10-22许可证:Meta Llama 3 Community
75Hammer2.1-1.5b (FC)MadeAgents开源27.983.069.515.60.00.0官方榜单2026-10-22许可证:cc-by-nc-4.0
76palmyra-x-004 (FC)Writer闭源27.987.577.90.42.513.1官方榜单2026-10-22许可证:Proprietary
77GPT-5-mini-2025-08-07 (Prompt)OpenAI闭源27.868.062.55.58.529.3官方榜单2026-10-22许可证:Proprietary
78Open-Mistral-Nemo-2407 (FC)Mistral AI闭源27.682.873.87.87.010.3官方榜单2026-10-22许可证:Proprietary
79GPT-5-nano-2025-08-07 (Prompt)OpenAI闭源27.680.870.70.813.524.5官方榜单2026-10-22许可证:Proprietary
80Amazon-Nova-2-Lite-v1:0 (FC)Amazon闭源27.187.080.82.15.02.4官方榜单2026-10-22许可证:Proprietary
80Granite-3.1-8B-Instruct (FC)IBM开源27.178.360.37.50.514.4官方榜单2026-10-22许可证:Apache-2.0
82Falcon3-10B-Instruct (FC)TII UAE开源27.085.075.46.51.527.5官方榜单2026-10-22许可证:falcon-llm-license
83Granite-3.2-8B-Instruct (FC)IBM开源26.979.860.37.40.512.5官方榜单2026-10-22许可证:Apache-2.0
84CoALM-8BUIUC + Oumi开源26.884.966.88.00.02.8官方榜单2026-10-22许可证:Meta Llama 3 Community
85Llama-3.1-8B-Instruct (Prompt)Meta开源25.884.070.811.13.010.8官方榜单2026-10-22许可证:Meta Llama 3 Community
86MiniCPM3-4B-FC (FC)openbmb开源25.681.865.23.90.012.0官方榜单2026-10-22许可证:Apache-2.0
87Claude-Haiku-4-5-20251001 (Prompt)Anthropic闭源25.355.452.51.819.52.6官方榜单2026-10-22许可证:Proprietary
88Amazon-Nova-Pro-v1:0 (FC)Amazon闭源25.086.678.51.92.51.9官方榜单2026-10-22许可证:Proprietary
89Claude-Sonnet-4-5-20250929 (Prompt)Anthropic闭源24.959.846.61.616.05.4官方榜单2026-10-22许可证:Proprietary
90GPT-4.1-nano-2025-04-14 (Prompt)OpenAI闭源24.972.450.32.01.516.8官方榜单2026-10-22许可证:Proprietary
91Falcon3-7B-Instruct (FC)TII UAE开源24.082.768.35.00.520.6官方榜单2026-10-22许可证:falcon-llm-license
92Qwen3-0.6B (FC)Qwen开源23.971.856.63.61.08.6官方榜单2026-10-22许可证:apache-2.0
93Granite-20b-FunctionCalling (FC)IBM开源23.282.358.75.40.00.0官方榜单2026-10-22许可证:Apache-2.0
94Qwen3-0.6B (Prompt)Qwen开源22.470.049.41.40.58.4官方榜单2026-10-22许可证:apache-2.0
95Amazon-Nova-Micro-v1:0 (FC)Amazon闭源22.374.166.31.41.52.4官方榜单2026-10-22许可证:Proprietary
96RZN-T (Prompt)Phronetic AI开源22.367.949.72.90.06.9官方榜单2026-10-22许可证:apache-2.0
97MiniCPM3-4B (Prompt)openbmb开源22.170.543.13.52.09.5官方榜单2026-10-22许可证:Apache-2.0
98Llama-3.2-3B-Instruct (FC)Meta开源21.982.758.34.01.06.2官方榜单2026-10-22许可证:Meta Llama 3 Community
99Bielik-11B-v2.3-Instruct (Prompt)SpeakLeash & ACK Cyfronet AGH开源21.981.567.82.61.511.4官方榜单2026-10-22许可证:Apache 2.0
100Hammer2.1-0.5b (FC)MadeAgents开源21.266.054.62.90.01.1官方榜单2026-10-22许可证:cc-by-nc-4.0
101Gemma-3-4b-it (Prompt)Google开源19.661.160.80.41.08.6官方榜单2026-10-22许可证:gemma-terms-of-use
102Open-Mistral-Nemo-2407 (Prompt)Mistral AI闭源19.388.574.00.82.58.6官方榜单2026-10-22许可证:Proprietary
103Granite-4.0-350m (FC)IBM开源19.067.946.12.50.53.2官方榜单2026-10-22许可证:Apache-2.0
104Falcon3-3B-Instruct (FC)TII UAE开源16.354.654.51.01.07.7官方榜单2026-10-22许可证:falcon-llm-license
105Ministral-8B-Instruct-2410 (FC)Mistral AI开源11.10.00.00.01.04.5官方榜单2026-10-22许可证:Mistral AI Research License
106Falcon3-1B-Instruct (FC)TII UAE开源11.19.02.90.00.05.8官方榜单2026-10-22许可证:falcon-llm-license
107Llama-3.2-1B-Instruct (FC)Meta开源10.838.411.80.00.03.2官方榜单2026-10-22许可证:Meta Llama 3 Community
108Llama-3.1-Nemotron-Ultra-253B-v1 (FC)NVIDIA开源10.00.00.00.00.00.0官方榜单2026-10-22许可证:nvidia-open-model-license
109Gemma-3-1b-it (Prompt)Google开源7.220.211.80.00.03.2官方榜单2026-10-22许可证:gemma-terms-of-use

榜单说明

第 4 批接入

Berkeley Function Calling Leaderboard(BFCL)由 UC Berkeley Gorilla 团队维护,评测大模型在可执行函数调用场景中的正确率。当前公开版本为 V4,在 v1–v3 的单轮/多轮调用之外,加入 Web Search、Memory 等 Agentic 子类,以及(对 Prompt 路线)格式敏感度相关设置。

榜单页面说明:总分 Overall Accuracy 为各子类分数的未加权平均;模型名中的 FC 表示原生 function/tool calling,Prompt 表示用普通文本生成模拟调用。页面标注最近更新为 2026-04-12,复现建议使用仓库 commit f7cf735 或 PyPI 包 bfcl-eval==2025.12.17。

本站每日同步官方页面加载的 data_overall.csv(与页面表格同源),收录全部模型;子项列为 CSV 中的 Non-Live、Live、Multi-Turn、Web Search、Memory 汇总准确率。许可证为 Proprietary 的记为闭源权重,其余具名许可证(含研究或非商用许可)记为开源权重。不同 FC/Prompt 设置为不同条目,不在此合并。评测代码与数据在 Gorilla 仓库的 berkeley-function-call-leaderboard 目录。

指标
  • Overall Accuracy:各评测子类准确率的未加权平均(%)(越高越好)
子项
单轮Non-LiveLive多轮Multi-TurnAgenticWeb SearchMemory
数据源
来源类型
官方榜单 109

分数取自数据源原文;同一模型在不同评测设置或不同来源下分别列出,来源冲突时全部保留并标注来源类型。

更新记录

0 条

首次同步作为基线,之后新上榜、分数变化和移除的条目会记录在这里。