V* Bench with Python
Multimodal · 2023-12-21
V* Bench with a code interpreter: the same 191 high-resolution visual-search questions answered while the model may run Python to crop, zoom or otherwise process the image. It is recorded separately from the no-tool V* Bench score.
Top models (higher is better)
| Model | Score |
|---|---|
| Qwen3.6 Plus (2026-04-02) | 96.9 |
| Qwen3.5 397B A17B | 95.8 |
| o3 | 95.7 |
| O4 Mini | 94.6 |
| Qwen3.5-27B | 93.7 |
| Qwen3.5 122B A10B | 93.2 |
| Qwen3.5 35B A3B | 92.7 |
| Qwen3.5-9B | 90.1 |
| Qwen3.5-4B | 84.3 |