Tool use & agents · %
BFCL v4 (function calling)
Berkeley Function Calling Leaderboard: single, parallel and multi-turn tool calls, web search, memory and relevance detection.
Active Contamination risk: Medium 107 results · 80 models Updated Oct 22, 2026 Newest result: 2026-10-17 Official leaderboard ↗
What it measures
Overall accuracy across categories; FC = native function calling, Prompt = tool calls via prompting.
Limitations
AST matching can mark valid alternative calls as wrong; leaderboard updates are periodic.
Contamination risk: Medium
Public test data.
Leaderboard
80 models
| # | Model | Provider | Score | Rank | Setting | Source type | Date |
|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 4.5 | Anthropic | 77.5% | 1 | Overall accuracy; native function calling (FC)“Claude-Opus-4-5-20251101 (FC)” | Official leaderboard | Oct 17, 2026 |
| 2 | Claude Sonnet 4.5 | Anthropic | 73.2% | 2 | Overall accuracy; native function calling (FC)“Claude-Sonnet-4-5-20250929 (FC)” | Official leaderboard | Oct 17, 2026 |
| 3 | Gemini 3 Pro | 72.5% | 3 | Overall accuracy; prompt-based“Gemini-3-Pro-Preview (Prompt)” | Official leaderboard | Oct 17, 2026 | |
| 4 | GLM-4.6OW | Zhipu AI (Z.ai) | 72.4% | 4 | Overall accuracy“GLM-4.6 (FC thinking)” | Official leaderboard | Oct 17, 2026 |
| 5 | Grok 4.1 fast | xAI | 69.6% | 5 | Overall accuracy; native function calling (FC)“Grok-4-1-fast-reasoning (FC)” | Official leaderboard | Oct 17, 2026 |
| 6 | Claude Haiku 4.5 | Anthropic | 68.7% | 6 | Overall accuracy; native function calling (FC)“Claude-Haiku-4-5-20251001 (FC)” | Official leaderboard | Oct 17, 2026 |
| 7 | o3 | OpenAI | 63.0% | 8 | Overall accuracy; prompt-based“o3-2025-04-16 (Prompt)” | Official leaderboard | Oct 17, 2026 |
| 8 | Grok 4 | xAI | 63.0% | 9 | Overall accuracy; prompt-based“Grok-4-0709 (Prompt)” | Official leaderboard | Oct 17, 2026 |
| 9 | Moonshotai-Kimi-K2-InstructOW | Moonshot AI | 59.1% | 11 | Overall accuracy; native function calling (FC)“Moonshotai-Kimi-K2-Instruct (FC)” | Official leaderboard | Oct 17, 2026 |
| 10 | Command A ReasoningOW | Cohere | 57.1% | 13 | Overall accuracy; native function calling (FC)“Command A Reasoning (FC)” | Official leaderboard | Oct 17, 2026 |
| 11 | DeepSeek-V3.2-Exp (Prompt + Thinking)OW | DeepSeek | 56.7% | 14 | Overall accuracy | Official leaderboard | Oct 17, 2026 |
| 12 | Gemini 2.5 Flash (Jun 2025) | 56.2% | 15 | Overall accuracy; native function calling (FC)“Gemini-2.5-Flash (FC)” | Official leaderboard | Oct 17, 2026 | |
| 13 | GPT-5.2 | OpenAI | 55.9% | 16 | Overall accuracy; native function calling (FC)“GPT-5.2-2025-12-11 (FC)” | Official leaderboard | Oct 17, 2026 |
| 14 | GPT-5 mini | OpenAI | 55.5% | 17 | Overall accuracy; native function calling (FC)“GPT-5-mini-2025-08-07 (FC)” | Official leaderboard | Oct 17, 2026 |
| 15 | xLAM-2-32b-fc-rOW | Salesforce | 54.7% | 18 | Overall accuracy; native function calling (FC)“xLAM-2-32b-fc-r (FC)” | Official leaderboard | Oct 17, 2026 |
| 16 | DeepSeek-V3.2-ExpOW | DeepSeek | 54.1% | 19 | Overall accuracy; native function calling (FC)“DeepSeek-V3.2-Exp (FC)” | Official leaderboard | Oct 17, 2026 |
| 17 | GPT-4.1 | OpenAI | 54.0% | 20 | Overall accuracy; native function calling (FC)“GPT-4.1-2025-04-14 (FC)” | Official leaderboard | Oct 17, 2026 |
| 18 | o4-mini | OpenAI | 53.2% | 21 | Overall accuracy; native function calling (FC)“o4-mini-2025-04-16 (FC)” | Official leaderboard | Oct 17, 2026 |
| 19 | xLAM-2-70b-fc-rOW | Salesforce | 53.1% | 22 | Overall accuracy; native function calling (FC)“xLAM-2-70b-fc-r (FC)” | Official leaderboard | Oct 17, 2026 |
| 20 | Qwen3-235B-A22B-Instruct (Jul 2025)OW | Alibaba (Qwen) | 52.1% | 23 | Overall accuracy; prompt-based“Qwen3-235B-A22B-Instruct-2507 (Prompt)” | Official leaderboard | Oct 17, 2026 |
| 21 | GPT-5 nano | OpenAI | 51.5% | 24 | Overall accuracy; native function calling (FC)“GPT-5-nano-2025-08-07 (FC)” | Official leaderboard | Oct 17, 2026 |
| 22 | Nanbeige4-3B-Thinking-2511OW | Nanbeige | 51.4% | 25 | Overall accuracy; native function calling (FC)“Nanbeige4-3B-Thinking-2511 (FC)” | Official leaderboard | Oct 17, 2026 |
| 23 | GPT-4.1 mini | OpenAI | 50.5% | 27 | Overall accuracy; native function calling (FC)“GPT-4.1-mini-2025-04-14 (FC)” | Official leaderboard | Oct 17, 2026 |
| 24 | Qwen3-32BOW | Alibaba (Qwen) | 48.7% | 29 | Overall accuracy; native function calling (FC)“Qwen3-32B (FC)” | Official leaderboard | Oct 17, 2026 |
| 25 | Nanbeige3.5-Pro-ThinkingOW | Nanbeige | 47.7% | 32 | Overall accuracy; native function calling (FC)“Nanbeige3.5-Pro-Thinking (FC)” | Official leaderboard | Oct 17, 2026 |
| 26 | xLAM-2-8b-fc-rOW | Salesforce | 46.7% | 34 | Overall accuracy; native function calling (FC)“xLAM-2-8b-fc-r (FC)” | Official leaderboard | Oct 17, 2026 |
| 27 | BitAgent-Bounty-8BOW | Bittensor | 46.2% | 36 | Overall accuracy | Official leaderboard | Oct 17, 2026 |
| 28 | Arch-Agent-32BOW | katanemo | 45.4% | 37 | Overall accuracy | Official leaderboard | Oct 17, 2026 |
| 29 | Qwen3-8BOW | Alibaba (Qwen) | 42.6% | 39 | Overall accuracy; native function calling (FC)“Qwen3-8B (FC)” | Official leaderboard | Oct 17, 2026 |
| 30 | ToolACE-2-8BOW | Huawei Noah & USTC | 42.4% | 40 | Overall accuracy; native function calling (FC)“ToolACE-2-8B (FC)” | Official leaderboard | Oct 17, 2026 |
| 31 | Qwen3-30B-A3B-Instruct (Jul 2025)OW | Alibaba (Qwen) | 41.4% | 41 | Overall accuracy; native function calling (FC)“Qwen3-30B-A3B-Instruct-2507 (FC)” | Official leaderboard | Oct 17, 2026 |
| 32 | xLAM-2-3b-fc-rOW | Salesforce | 41.2% | 42 | Overall accuracy; native function calling (FC)“xLAM-2-3b-fc-r (FC)” | Official leaderboard | Oct 17, 2026 |
| 33 | Qwen3-14BOW | Alibaba (Qwen) | 41.0% | 43 | Overall accuracy; native function calling (FC)“Qwen3-14B (FC)” | Official leaderboard | Oct 17, 2026 |
| 34 | Mistral Large 2 (Nov 2024)OW | Mistral AI | 38.4% | 46 | Overall accuracy; native function calling (FC)“mistral-large-2411 (FC)” | Official leaderboard | Oct 17, 2026 |
| 35 | Mistral Medium 3 | Mistral AI | 37.7% | 48 | Overall accuracy“Mistral-Medium-2505” | Official leaderboard | Oct 17, 2026 |
| 36 | Llama 4 MaverickOW | Meta | 37.3% | 50 | Overall accuracy; native function calling (FC)“Llama-4-Maverick-17B-128E-Instruct-FP8 (FC)” | Official leaderboard | Oct 17, 2026 |
| 37 | Mistral Small 3.2OW | Mistral AI | 37.1% | 51 | Overall accuracy; native function calling (FC)“Mistral-small-2506 (FC)” | Official leaderboard | Oct 17, 2026 |
| 38 | Gemini-2.5-Flash-Lite | 36.9% | 52 | Overall accuracy; native function calling (FC)“Gemini-2.5-Flash-Lite (FC)” | Official leaderboard | Oct 17, 2026 | |
| 39 | qwen3-4b-instruct-2507OW | Alibaba (Qwen) | 35.7% | 54 | Overall accuracy; native function calling (FC)“Qwen3-4B-Instruct-2507 (FC)” | Official leaderboard | Oct 17, 2026 |
| 40 | Arch-Agent-3BOW | katanemo | 35.4% | 56 | Overall accuracy | Official leaderboard | Oct 17, 2026 |
| 41 | GPT-4.1 nano | OpenAI | 33.0% | 58 | Overall accuracy; native function calling (FC)“GPT-4.1-nano-2025-04-14 (FC)” | Official leaderboard | Oct 17, 2026 |
| 42 | Arch-Agent-1.5BOW | katanemo | 32.1% | 60 | Overall accuracy | Official leaderboard | Oct 17, 2026 |
| 43 | Command R7BOW | Cohere | 32.1% | 61 | Overall accuracy; native function calling (FC)“Command R7B (FC)” | Official leaderboard | Oct 17, 2026 |
| 44 | Llama 3.3 70BOW | Meta | 31.9% | 62 | Overall accuracy; native function calling (FC)“Llama-3.3-70B-Instruct (FC)” | Official leaderboard | Oct 17, 2026 |
| 45 | Hammer2.1-7bOW | MadeAgents | 31.7% | 64 | Overall accuracy; native function calling (FC)“Hammer2.1-7b (FC)” | Official leaderboard | Oct 17, 2026 |
| 46 | xLAM-2-1b-fc-rOW | Salesforce | 30.4% | 65 | Overall accuracy; native function calling (FC)“xLAM-2-1b-fc-r (FC)” | Official leaderboard | Oct 17, 2026 |
| 47 | Gemma 3 12BOW | 30.4% | 66 | Overall accuracy; prompt-based“Gemma-3-12b-it (Prompt)” | Official leaderboard | Oct 17, 2026 | |
| 48 | Hammer2.1-3bOW | MadeAgents | 29.7% | 68 | Overall accuracy; native function calling (FC)“Hammer2.1-3b (FC)” | Official leaderboard | Oct 17, 2026 |
| 49 | Gemma 3 27BOW | 29.5% | 69 | Overall accuracy; prompt-based“Gemma-3-27b-it (Prompt)” | Official leaderboard | Oct 17, 2026 | |
| 50 | Phi-4OW | Microsoft Research | 28.8% | 70 | Overall accuracy; prompt-based“Phi-4 (Prompt)” | Official leaderboard | Oct 17, 2026 |
| 51 | Qwen3-1.7BOW | Alibaba (Qwen) | 28.4% | 71 | Overall accuracy; native function calling (FC)“Qwen3-1.7B (FC)” | Official leaderboard | Oct 17, 2026 |
| 52 | Llama 4 ScoutOW | Meta | 28.1% | 72 | Overall accuracy; native function calling (FC)“Llama-4-Scout-17B-16E-Instruct (FC)” | Official leaderboard | Oct 17, 2026 |
| 53 | CoALM-70BOW | UIUC + Oumi | 28.0% | 74 | Overall accuracy | Official leaderboard | Oct 17, 2026 |
| 54 | Hammer2.1-1.5bOW | MadeAgents | 27.9% | 75 | Overall accuracy; native function calling (FC)“Hammer2.1-1.5b (FC)” | Official leaderboard | Oct 17, 2026 |
| 55 | Palmyra-X-004 | Writer | 27.9% | 76 | Overall accuracy; native function calling (FC)“palmyra-x-004 (FC)” | Official leaderboard | Oct 17, 2026 |
| 56 | Mistral NeMoOW | Mistral AI | 27.6% | 78 | Overall accuracy; native function calling (FC)“Open-Mistral-Nemo-2407 (FC)” | Official leaderboard | Oct 17, 2026 |
| 57 | Amazon-Nova-2-Lite-v1:0 | Amazon | 27.1% | 80 | Overall accuracy; native function calling (FC)“Amazon-Nova-2-Lite-v1:0 (FC)” | Official leaderboard | Oct 17, 2026 |
| 58 | Granite-3.1-8B-InstructOW | IBM | 27.1% | 81 | Overall accuracy; native function calling (FC)“Granite-3.1-8B-Instruct (FC)” | Official leaderboard | Oct 17, 2026 |
| 59 | Falcon3-10B-InstructOW | TII UAE | 27.0% | 82 | Overall accuracy; native function calling (FC)“Falcon3-10B-Instruct (FC)” | Official leaderboard | Oct 17, 2026 |
| 60 | Granite-3.2-8B-InstructOW | IBM | 26.9% | 83 | Overall accuracy; native function calling (FC)“Granite-3.2-8B-Instruct (FC)” | Official leaderboard | Oct 17, 2026 |
| 61 | CoALM-8BOW | UIUC + Oumi | 26.8% | 84 | Overall accuracy | Official leaderboard | Oct 17, 2026 |
| 62 | Llama 3.1-8BOW | Meta | 25.8% | 85 | Overall accuracy; prompt-based“Llama-3.1-8B-Instruct (Prompt)” | Official leaderboard | Oct 17, 2026 |
| 63 | MiniCPM3-4B-FCOW | openbmb | 25.6% | 86 | Overall accuracy; native function calling (FC)“MiniCPM3-4B-FC (FC)” | Official leaderboard | Oct 17, 2026 |
| 64 | Amazon Nova Pro | Amazon | 25.0% | 88 | Overall accuracy; native function calling (FC)“Amazon-Nova-Pro-v1:0 (FC)” | Official leaderboard | Oct 17, 2026 |
| 65 | Falcon3-7B-InstructOW | TII UAE | 24.0% | 91 | Overall accuracy; native function calling (FC)“Falcon3-7B-Instruct (FC)” | Official leaderboard | Oct 17, 2026 |
| 66 | Qwen3-0.6BOW | Alibaba (Qwen) | 23.9% | 92 | Overall accuracy; native function calling (FC)“Qwen3-0.6B (FC)” | Official leaderboard | Oct 17, 2026 |
| 67 | Granite-20b-FunctionCallingOW | IBM | 23.2% | 93 | Overall accuracy; native function calling (FC)“Granite-20b-FunctionCalling (FC)” | Official leaderboard | Oct 17, 2026 |
| 68 | Amazon Nova Micro | Amazon | 22.3% | 95 | Overall accuracy; native function calling (FC)“Amazon-Nova-Micro-v1:0 (FC)” | Official leaderboard | Oct 17, 2026 |
| 69 | RZN-TOW | Phronetic AI | 22.3% | 96 | Overall accuracy; prompt-based“RZN-T (Prompt)” | Official leaderboard | Oct 17, 2026 |
| 70 | Llama-3.2-3B-InstructOW | Meta | 21.9% | 98 | Overall accuracy; native function calling (FC)“Llama-3.2-3B-Instruct (FC)” | Official leaderboard | Oct 17, 2026 |
| 71 | Bielik-11B-v2.3-InstructOW | SpeakLeash & ACK Cyfronet AGH | 21.9% | 99 | Overall accuracy; prompt-based“Bielik-11B-v2.3-Instruct (Prompt)” | Official leaderboard | Oct 17, 2026 |
| 72 | Hammer2.1-0.5bOW | MadeAgents | 21.2% | 100 | Overall accuracy; native function calling (FC)“Hammer2.1-0.5b (FC)” | Official leaderboard | Oct 17, 2026 |
| 73 | Gemma 3 4BOW | 19.6% | 101 | Overall accuracy; prompt-based“Gemma-3-4b-it (Prompt)” | Official leaderboard | Oct 17, 2026 | |
| 74 | granite-4.0-350mOW | IBM | 19.0% | 103 | Overall accuracy; native function calling (FC)“Granite-4.0-350m (FC)” | Official leaderboard | Oct 17, 2026 |
| 75 | Falcon3-3B-InstructOW | TII UAE | 16.3% | 104 | Overall accuracy; native function calling (FC)“Falcon3-3B-Instruct (FC)” | Official leaderboard | Oct 17, 2026 |
| 76 | Ministral 8BOW | Mistral AI | 11.1% | 105 | Overall accuracy; native function calling (FC)“Ministral-8B-Instruct-2410 (FC)” | Official leaderboard | Oct 17, 2026 |
| 77 | Falcon3-1B-InstructOW | TII UAE | 11.1% | 106 | Overall accuracy; native function calling (FC)“Falcon3-1B-Instruct (FC)” | Official leaderboard | Oct 17, 2026 |
| 78 | Llama 3.2 1BOW | Meta | 10.8% | 107 | Overall accuracy; native function calling (FC)“Llama-3.2-1B-Instruct (FC)” | Official leaderboard | Oct 17, 2026 |
| 79 | Llama-3.1-Nemotron-Ultra-253B-v1OW | NVIDIA | 10.0% | 108 | Overall accuracy; native function calling (FC)“Llama-3.1-Nemotron-Ultra-253B-v1 (FC)” | Official leaderboard | Oct 17, 2026 |
| 80 | Gemma 3 1BOW | 7.2% | 109 | Overall accuracy; prompt-based“Gemma-3-1b-it (Prompt)” | Official leaderboard | Oct 17, 2026 |
Data sources & licences
| Source | Licence / terms | How we reuse it | Last fetched |
|---|---|---|---|
| Berkeley Function Calling Leaderboard (BFCL) Gorilla / BFCL team, UC Berkeley | Apache-2.0 (gorilla repo) | Open licence – full data with attribution | Oct 22, 2026 |
FAQ
What does BFCL v4 (function calling) measure?
Berkeley Function Calling Leaderboard: single, parallel and multi-turn tool calls, web search, memory and relevance detection. Overall accuracy across categories; FC = native function calling, Prompt = tool calls via prompting.
How reliable is BFCL v4 (function calling)?
AST matching can mark valid alternative calls as wrong; leaderboard updates are periodic. Contamination risk: Medium – Public test data.