Visual Question Answering on DV-500
90.2Accuracy (Original)Qwen2.5-VL
Evaluation Results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Qwen2.5-VL#Params=7B, #Tokens=2N, Finetuning Status=w/o Finetuning2026.04 | 90.2 | 75 | 65 | 61 | 60.6 | 89.4 | 83.8 | 80.6 | 72.8 | 85.8 | 78.2 | 71.4 | 61.2 | |
| Qwen2-VL#Params=7B, #Tokens=2N, Finetuning Status=w/o Finetuning2026.04 | 89.8 | 77.8 | 73.6 | 70.8 | 69.4 | 89.6 | 85.6 | 82.6 | 78.4 | 85 | 79.4 | 75.2 | 65.6 | |
| LLaVA-Next Interleave Qwen#Params=7B, #Tokens=2N, Finetuning Status=w/o Finetuning2026.04 | 88.6 | 81.4 | 78.4 | 75.2 | 73.6 | 86.4 | 86 | 85.6 | 81.4 | 85 | 83.8 | 79.8 | 73.4 | |
| DUALVISION#Params=7B, #Tokens=N, Finetuning Status=Finetuned2026.04 | 88.38 | 84.77 | 84.37 | 82.36 | 82.57 | 88.38 | 87.58 | 87.58 | 86.57 | 87.98 | 87.17 | 84.97 | 80.76 | |
| Claude Sonnet-3.5 v2#Params=–, #Tokens=2N, Finetuning Status=w/o Finetuning2026.04 | 87.4 | 77.8 | 74.8 | 70.8 | 72 | 85.4 | 78.8 | 75.6 | 68 | 80 | 70.2 | 68.4 | 64.4 | |
| LLaVA 1.5#Params=7B, #Tokens=2N, Finetuning Status=Finetuned2026.04 | 87.15 | 83.5 | 82.13 | 81.93 | 80.32 | 86.35 | 85.94 | 85.54 | 84.94 | 85.94 | 85.94 | 84.74 | 78.71 | |
| Claude Opus 4#Params=–, #Tokens=2N, Finetuning Status=w/o Finetuning2026.04 | 86.6 | 67 | 60.8 | 60 | 58.8 | 83.6 | 72.4 | 64.6 | 57 | 74.2 | 63.8 | 59.2 | 56.2 | |
| LLaMA-4 Scout#Params=17B, #Tokens=2N, Finetuning Status=w/o Finetuning2026.04 | 85.8 | 71.6 | 62.6 | 59.2 | 57.6 | 79.4 | 71.6 | 66.4 | 56.6 | 73.4 | 63.6 | 58 | 54 | |
| LLaVA 1.5#Params=7B, #Tokens=2N, Finetuning Status=w/o Finetuning2026.04 | 81.2 | 79 | 76 | 73.2 | 72.6 | 74 | 72.8 | 73.2 | 71.2 | 81 | 79.2 | 78.4 | 71.4 | |
| LLaVA 1.5–Text Only#Params=7B, #Tokens=–, Finetuning Status=w/o Finetuning2026.04 | 48.8 | — | — | — | — | — | — | — | — | — | — | — | — |