BFCL v4 Overall Accuracy
Agents · 2025-07-17
Overall accuracy on the BFCL v4 function-calling benchmark.
Top models (higher is better)
| Model | Score |
|---|---|
| Opus 4.5 | 77.5 |
| Atria Dawn Preview | 77.0 |
| Sonnet 4.5 | 73.2 |
| Gemini 3 Pro Preview | 72.5 |
| GLM 4.6 | 72.4 |
| Grok 4.1 Fast | 69.6 |
| Haiku 4.5 | 68.7 |
| o3 | 63.0 |
| Grok 4 | 63.0 |
| Kimi K2 Instruct | 59.1 |
| Command A Reasoning | 57.1 |
| DeepSeek-V3.2-Exp | 56.7 |
| Gemini 2.5 Flash | 56.2 |
| GPT-5.2 | 55.9 |
| GPT-5 Mini | 55.5 |
| GPT-4.1 | 54.0 |
| O4 Mini | 53.2 |
| Qwen3 235B A22B Instruct 2507 | 52.1 |
| GPT-5 Nano | 51.5 |
| Nanbeige4-3B-Thinking-2511 | 51.4 |
| GPT-4.1 Mini | 50.5 |
| Qwen3 32B | 48.7 |
| Command A | 46.5 |
| Qwen3 8B | 42.6 |
| Qwen3 30B A3B Instruct 2507 | 41.4 |
| Qwen3 14B | 41.0 |
| Mistral Large 2.1 (Instruct 2411) | 38.4 |
| Mistral Medium 3 | 37.7 |
| Llama 4 Maverick Instruct FP8 | 37.3 |
| Mistral Small 3.2 24B Instruct 2506 | 37.1 |
| Gemini 2.5 Flash-Lite | 36.9 |
| Qwen3 4B Instruct 2507 | 35.7 |
| GPT-4.1 Nano | 33.0 |
| Command R7B (Dec 2024) | 32.1 |
| Llama 3.3 70B Instruct | 31.9 |
| Gemma 3 12B IT | 30.4 |
| Gemma 3 27B IT | 29.5 |
| Phi-4 | 28.8 |
| Qwen3 1.7B | 28.4 |
| Llama 4 Scout Instruct | 28.1 |