Computer Vision Benchmarking on CVBench
83AccuracyAutoV
Evaluation results
| Method | Links | |
|---|---|---|
| AutoVModel=LLaVA-OneVision 7B2025.06 | 83 | |
| AutoVModel=Qwen2.5-VL 7B2025.06 | 83 | |
| APIModel=Qwen2.5-VL 7B2025.06 | 81.1 | |
| APIModel=LLaVA-OneVision 7B2025.06 | 80.7 | |
| BaseModel=Qwen2.5-VL 7B2025.06 | 80.6 | |
| BaseModel=LLaVA-OneVision 7B2025.06 | 79.6 | |
| LLaVA-665K# Data=665K2025.04 | 60.92 | |
| MSA-PTTraining Strategy=from-scratch sparse pretraining2026.06 | 59.7 | |
| MSA-CPTTraining Strategy=sparse continued pretraining2026.06 | 58.8 | |
| FullTraining Strategy=Full-Attention baseline2026.06 | 57 | |
| ICONS# Data=65K2025.04 | 55.96 | |
| COMPACT# Data=65K2025.04 | 55.28 | |
| Random# Data=65K2025.04 | 54.71 | |
| Self-Sup# Data=65K2025.04 | 54.3 | |
| Perplexity# Data=65K2025.04 | 52.72 | |
| EL2N# Data=65K2025.04 | 50.92 | |
| D2-Pruning# Data=65K2025.04 | 48.49 | |
| Self-Filter# Data=65K2025.04 | 45.17 | |
| SemDeDup# Data=65K2025.04 | 42.24 |