Visual Tool-Use on HRBench 8K
73.7AccuracyARM-Thinker-7B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ARM-Thinker-7BSize=7B, Backbone=Qwen2.5-VL-7B2025.12 | 73.7 | — | |
| Mini-o3Source=Lai et al. (2025)2025.12 | 73.3 | — | |
| CodeV-7B-RLBackbone=7B, Training=RL2025.11 | 71.3 | 13.3 | |
| Thyme-RL-7BBackbone=7B, Training=RL2025.11 | 71.2 | 1.2 | |
| Qwen3-VL-8BSize=8B2025.12 | 70.4 | — | |
| InternVL3.5-8BSize=8B2025.12 | 69.9 | — | |
| DeepEyesSource=Lai et al. (2025)2025.12 | 69.5 | — | |
| DeepEyes-7BBackbone=7B2025.11 | 69.1 | 6.7 | |
| Pixel-Reasoner-7BBackbone=7B2025.11 | 68.5 | 7.6 | |
| InternVL3-8BSize=8B2025.12 | 68.4 | — | |
| Pixel ReasonerSource=Lai et al. (2025)2025.12 | 66.9 | — | |
| Qwen2.5-VL-7BSize=7B2025.12 | 64.6 | — | |
| GPT-4oSource=Lai et al. (2025)2025.12 | 58.3 | — |