GTA — leaderboard
General Tool Agents benchmark from OpenCompass. Tests LLMs on multi-step tool use across perception, operation, logic, and creativity categories using real-world tool chains.
Metric: Answer Accuracy (%). Source: github.com. Status: saturation imminent. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 45.07 |
| 2 | DeepSeek V3.2 | 42.68 |
| 3 | Qwen 3 235B A22B | 42.21 |
| 4 | Gemini 2.5 Pro | 39.58 |
| 5 | Qwen 3 8B | 27.1 |
| 6 | Llama 3.2 3B Instruct | 24.99 |
| 7 | Llama 4 Scout | 22.03 |
| 8 | Kimi K2 | 18.32 |
| 9 | Claude Sonnet 4.5 | 17.89 |
| 10 | Grok 4 | 17.78 |
| 11 | Qwen 3 30B A3B | 15.67 |
| 12 | Llama 3.1 70B Instruct | 13.24 |
| 13 | Llama 3.1 8B Instruct | 8.78 |
Interactive version: theaggregate.ai/benchmark?slug=gta · How the rankings work · Data refreshed daily, snapshot 2026-07-22.