Berkeley Function Calling Leaderboard — evaluates LLM ability to correctly generate function/tool calls across simple, parallel, multiple, and multi-turn scenarios.
Tests real-world tool use with diverse calling patterns; includes live, multi-turn, web search, and memory categories.
Models appear as FC (native function calling) and Prompt variants; we take the best score.
Higher is better. Percentage of correctly generated function calls.
Models appear as FC (native function calling) and Prompt variants.
| # | Model | Score |
|---|---|---|
| 1 | 77.5 | |
| 2 | 77.5 | |
| 3 | 73.2 | |
| 4 | 73.2 | |
| 5 | 72.5 | |
| 6 | 72.5 | |
| 7 | 72.4 | |
| 8 | 72.4 | |
| 9 | 69.6 | |
| 10 | 68.7 | |
| 11 | 68.7 | |
| 12 | 66.7 | |
| 13 | 66.6 | |
| 14 | 66.0 | |
| 15 | 63.0 | |
| 16 | 63.0 | |
| 17 | 62.9 | |
| 18 | 62.7 | |
| 19 | 62.0 | |
| 20 | 61.8 | |
| 21 | 61.3 | |
| 22 | 61.0 | |
| 23 | 60.7 | |
| 24 | 60.4 | |
| 25 | 60.3 | |
| 26 | 60.1 | |
| 27 | 59.8 | |
| 28 | 59.6 | |
| 29 | 59.4 | |
| 30 | 59.3 | |
| 31 | 59.1 | |
| 32 | 59.0 | |
| 33 | 58.6 | |
| 34 | 58.4 | |
| 35 | 58.3 | |
| 36 | 58.3 | |
| 37 | 58.3 | |
| 38 | 58.3 | |
| 39 | 58.2 | |
| 40 | 57.9 | |
| 41 | 57.8 | |
| 42 | 57.7 | |
| 43 | 57.5 | |
| 44 | 57.5 | |
| 45 | 57.1 | |
| 46 | 57.0 | |
| 47 | 56.9 | |
| 48 | 56.8 | |
| 49 | 56.8 | |
| 50 | 56.7 |