ToolBench - Tabletop: leaderboard
ToolBench subtask measuring tabletop-world action-generation success.
Metric: Task Score. Source: huggingface.co. Saturation forecast: Estimated already saturated. 44 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 | 81 | #438 |
| 2 | text-davinci-003 | 66.7 | #1146 |
| 3 | CodeLlama-13B-Instruct-hf | 57.62 | #1532 |
| 4 | CodeLlama-34B-hf | 51.32 | #1515 |
| 5 | CodeLlama-34B-Instruct-hf | 47.46 | #1381 |
| 6 | Llama 2 70B | 45.4 | #1201 |
| 7 | CodeLlama-13B-Python-hf | 41.53 | #1623 |
| 8 | LLaMA-30B | 34.3 | #1423 |
| 9 | CodeLlama-34B-Python-hf | 33.33 | #1552 |
| 10 | GPT-3.5 Turbo | 33.3 | #865 |
| 11 | LLaMA-65B | 30.5 | #1303 |
| 12 | Llama 2 13B | 23.81 | #1436 |
| 13 | CodeLlama-7B-Python-hf | 22.86 | #1608 |
| 14 | starcoder | 21.9 | #1577 |
| 15 | LLaMA-13B | 17.1 | #1516 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=toolbench-tabletop · How It Works · Data refreshed daily, snapshot 2026-10-20.