benchmark evidence
BFCL Live
BFCL live function calling accuracy.
winner on BFCL Live
direct benchmark result, not a broad vertical composite | source row dated 2026-05-15
scored on 2026-05-15
latest mapped results | top 20
| # | Model | Score | Evidence | Tested |
|---|---|---|---|---|
| 1 | Mistral: Mistral Small 4 | 79.0 | model-only independent_benchmark | 2026-05-15 |
| 2 | OpenAI: GPT-4.1 | 78.9 | model-only independent_benchmark | 2026-05-15 |
| 3 | MoonshotAI: Kimi K2 Thinking | 78.7 | model-only independent_benchmark | 2026-05-15 |
| 4 | Amazon: Nova Pro 1.0 | 78.5 | model-only independent_benchmark | 2026-05-15 |
| 5 | Google: Gemini 2.5 Flash | 78.2 | model-only independent_benchmark | 2026-05-15 |
| 6 | Meta: Llama 3.3 70B Instruct | 76.6 | model-only independent_benchmark | 2026-05-15 |
| 7 | Anthropic: Claude Opus 4.5 | 76.0 | model-only independent_benchmark | 2026-05-15 |
| 8 | Meta: Llama 4 Scout | 74.7 | model-only independent_benchmark | 2026-05-15 |
| 9 | Google: Gemma 3 27B | 74.5 | model-only independent_benchmark | 2026-05-15 |
| 10 | Meta: Llama 4 Maverick | 73.7 | model-only independent_benchmark | 2026-05-15 |
| 11 | OpenAI: o4 Mini | 70.8 | model-only independent_benchmark | 2026-05-15 |
| 12 | Mistral Large | 68.1 | model-only independent_benchmark | 2026-05-15 |
| 13 | OpenAI: GPT-5.2 | 67.1 | model-only independent_benchmark | 2026-05-15 |
| 14 | OpenAI: o3 | 66.2 | model-only independent_benchmark | 2026-05-15 |
| 15 | DeepSeek: DeepSeek V3.2 | 53.7 | model-only independent_benchmark | 2026-05-15 |
| 16 | Anthropic: Claude Haiku 4.5 | 52.5 | model-only independent_benchmark | 2026-05-15 |
| 17 | Anthropic: Claude Sonnet 4.5 | 46.6 | model-only independent_benchmark | 2026-05-15 |
what this result means
BFCL live function calling accuracy.
This benchmark contributes direct public evidence. Read its scope before generalizing the result.
A win here is a win on BFCL Live. Broad task pages require independent corroboration before naming a general winner.
source record
category: structured_output
metric: accuracy
matched models: 17
latest source date: 2026-05-15
direction: higher is better