Multi-image understanding on QBench2
81.7AccuracyDelimScaling
Evaluation results
| Method | Links | |
|---|---|---|
| DelimScalingBackbone=Qwen2.5-VL, Model Size=32B2026.02 | 81.7 | |
| Qwen2.5-VLModel Size=32B, Method Variant=Baseline2026.02 | 81.4 | |
| DelimScalingBackbone=InternVL3, Model Size=14B2026.02 | 80.1 | |
| InternVL3Model Size=14B, Method Variant=Baseline2026.02 | 79.6 | |
| Qwen2VLParams=7B2026.03 | 76.8 | |
| DelimScalingBackbone=InternVL3, Model Size=8B2026.02 | 76.6 | |
| DelimScalingBackbone=Qwen2.5-VL, Model Size=7B2026.02 | 76.5 | |
| InternVL3Model Size=8B, Method Variant=Baseline2026.02 | 76.5 | |
| Qwen2.5-VLModel Size=7B, Method Variant=Baseline2026.02 | 75.8 | |
| InternVL2.5Params=8B2026.03 | 75.5 | |
| InternVL2.5 + CAPLParams=8B2026.03 | 75.3 | |
| GLM4.1VBaseParams=9B2026.03 | 74.4 | |
| DelimScalingBackbone=LLaVA-OV, Model Size=7B2026.02 | 74.2 | |
| LLaVA-OVModel Size=7B, Method Variant=Baseline2026.02 | 73.9 | |
| GLM4.1VBase + CAPLParams=9B2026.03 | 73.6 | |
| Qwen2.5-VL + CAPLParams=7B2026.03 | 72.4 | |
| Qwen2.5-VLParams=7B2026.03 | 71.1 | |
| LLaVA-OVParams=7B2026.03 | 70.1 | |
| InternVL2Params=7B2026.03 | 69.8 | |
| DelimScalingBackbone=InternVL3, Model Size=2B2026.02 | 65.6 | |
| InternVL3Model Size=2B, Method Variant=Baseline2026.02 | 65.2 | |
| Idefics3Params=8B2026.03 | 64.2 | |
| DelimScalingBackbone=Qwen2.5-VL, Model Size=3B2026.02 | 63.3 | |
| Qwen2.5-VLModel Size=3B, Method Variant=Baseline2026.02 | 62.7 | |
| DelimScalingBackbone=LLaVA-OV, Model Size=0.5B2026.02 | 51.9 | |
| LLaVA-OVModel Size=0.5B, Method Variant=Baseline2026.02 | 51.7 | |
| InternVL3Model Size=1B, Method Variant=Baseline2026.02 | 50.8 | |
| DelimScalingBackbone=InternVL3, Model Size=1B2026.02 | 50.2 | |
| LLaVA-NextParams=7B2026.03 | 39.5 | |
| Idefics2Params=8B2026.03 | 38.6 |