Multi-image Visual Question Answering on Mantis
76.5AccuracyDelimScaling
Evaluation Results
| Method | Links | |
|---|---|---|
| DelimScalingBackbone=InternVL3, Model Size=78B2026.02 | 76.5 | |
| DelimScalingBackbone=Qwen2.5-VL, Model Size=72B2026.02 | 75.58 | |
| BaselineBackbone=InternVL3, Model Size=78B2026.02 | 74.65 | |
| BaselineBackbone=Qwen2.5-VL, Model Size=72B2026.02 | 74.19 |