ToolBench Leaderboard — leaderboard
ToolBench action-generation leaderboard measuring how well models manipulate APIs and tools across weather, search, booking, spreadsheet, WebShop, and tabletop tasks.
Metric: Average Task Score. Source: huggingface.co. Status: saturation imminent. 44 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-4 | 68.76 |
| 2 | text-davinci-003 | 67.24 |
| 3 | CodeLlama-34B-Instruct-hf | 64.79 |
| 4 | CodeLlama-34B-hf | 62.94 |
| 5 | Llama 2 70B | 61.03 |
| 6 | CodeLlama-13B-Instruct-hf | 60.49 |
| 7 | CodeLlama-34B-Python-hf | 59.16 |
| 8 | CodeLlama-13B-hf | 56.89 |
| 9 | GPT-3.5 Turbo | 56.65 |
| 10 | CodeLlama-13B-Python-hf | 56.31 |
| 11 | LLaMA-65B | 55.59 |
| 12 | CodeLlama-7B-Python-hf | 52.15 |
| 13 | CodeLlama-7B-Instruct-hf | 50.48 |
| 14 | starcoder | 49.75 |
| 15 | LLaMA-30B | 49.59 |
Interactive version: theaggregate.ai/benchmark?slug=toolbench-leaderboard · How the rankings work · Data refreshed daily, snapshot 2026-07-22.