VisualToolBench (VTB)
Agents · 2025-10-15
Score on multimodal tool-use tasks that require manipulating and reasoning over images.
Top models (higher is better)
| Model | Score |
|---|---|
| Muse Spark 1.1 | 44.8 |
| GPT-5.4 | 29.2 |
| Gemini 3.1 Pro Preview | 29.0 |
| Opus 4.6 | 27.5 |
| Gemini 3 Pro Preview | 26.9 |
| GPT-5 | 18.7 |
| o3 | 13.7 |
| Gemini 2.5 Pro Preview 06-05 | 11.8 |
| O4 Mini | 11.1 |
| Sonnet 4.5 | 6.2 |
| GPT-4.1 | 5.5 |
| Opus 4.1 | 5.2 |
| Gemini 2.5 Flash | 4.7 |
| Sonnet 4 | 4.5 |
| Nova Premier | 2.0 |
| Llama 4 Scout Instruct | 1.6 |
| Llama 4 Maverick Instruct | 1.4 |