Image-text Understanding on MMBench
91.22AccuracyDAPO + ReMind
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DAPO + ReMindBase Model=Qwen3-VL-8B-Instruct2026.06 | 91.22 | — | |
| DAPOBase Model=Qwen3-VL-8B-Instruct2026.06 | 90.53 | — | |
| RLEPBase Model=Qwen3-VL-8B-Instruct2026.06 | 90.45 | — | |
| RePOBase Model=Qwen3-VL-8B-Instruct2026.06 | 90.45 | — | |
| ExGRPOBase Model=Qwen3-VL-8B-Instruct2026.06 | 90.44 | — | |
| GRPO + ReMindBase Model=Qwen3-VL-8B-Instruct2026.06 | 90.41 | — | |
| GRPOBase Model=Qwen3-VL-8B-Instruct2026.06 | 90.29 | — | |
| Base ModelBase Model=Qwen3-VL-8B-Instruct2026.06 | 89.79 | — | |
| PerceptioParameters=8B2026.03 | 83.4 | — | |
| Qwen2.5-VL 7BToken Budget=All 1296 Tokens, Reduction Percentage=0%2026.03 | 82.9 | 100 | |
| VisPrunerToken Budget=512, Reduction Percentage=60.5%2026.03 | 82.6 | 94 | |
| Sa2VAParameters=8B2026.03 | 82.4 | — | |
| DivPruneToken Budget=512, Reduction Percentage=60.5%2026.03 | 82.2 | 94.2 | |
| PromPruneToken Budget=512, Reduction Percentage=60.5%2026.03 | 82.1 | 94.8 | |
| PerceptioParameters=4B2026.03 | 82 | — | |
| InternVL2Parameters=8B2026.03 | 81.7 | — | |
| VisPrunerToken Budget=256, Reduction Percentage=80.2%2026.03 | 80.5 | 91.7 | |
| PromPruneToken Budget=256, Reduction Percentage=80.2%2026.03 | 80.3 | 92.6 | |
| DivPruneToken Budget=256, Reduction Percentage=80.2%2026.03 | 79.7 | 91.1 | |
| mPLUG-Owl3Parameters=8B2026.03 | 77.6 | — | |
| DivPruneToken Budget=128, Reduction Percentage=90.1%2026.03 | 76.8 | 86.5 | |
| Sa2VAParameters=4B2026.03 | 76.8 | — | |
| VisPrunerToken Budget=128, Reduction Percentage=90.1%2026.03 | 76.6 | 86.2 | |
| PromPruneToken Budget=128, Reduction Percentage=90.1%2026.03 | 76.5 | 87.9 | |
| LLAVA-1.5Parameters=13B2026.03 | 68.8 | — | |
| VisPrunerToken Budget=640, Base Model=LLaVA-NeXT 7B2026.03 | 65.5 | 99.2 | |
| DivPruneToken Budget=640, Base Model=LLaVA-NeXT 7B2026.03 | 65.5 | 99.6 | |
| PromPruneToken Budget=640, Base Model=LLaVA-NeXT 7B2026.03 | 65 | 99.6 | |
| LLaVA-NeXT 7BToken Budget=2880, Base Model=LLaVA-NeXT 7B2026.03 | 63.9 | 100 | |
| VisPrunerToken Budget=320, Base Model=LLaVA-NeXT 7B2026.03 | 63.4 | 95.9 | |
| DivPruneToken Budget=320, Base Model=LLaVA-NeXT 7B2026.03 | 63.4 | 97.7 | |
| PromPruneToken Budget=320, Base Model=LLaVA-NeXT 7B2026.03 | 63.1 | 97.9 | |
| PromPruneToken Budget=160, Base Model=LLaVA-NeXT 7B2026.03 | 62.3 | 95.2 | |
| DivPruneToken Budget=160, Base Model=LLaVA-NeXT 7B2026.03 | 61.6 | 95 | |
| Video-LLaVAParameters=7B2026.03 | 60.9 | — | |
| VisPrunerToken Budget=160, Base Model=LLaVA-NeXT 7B2026.03 | 58.4 | 90.6 | |
| OMG-LLaVAParameters=7B2026.03 | 47.9 | — | |
| GLaMMParameters=7B2026.03 | 36.8 | — | |
| PixelLMParameters=7B2026.03 | 17.4 | — | |
| LaSagnA2026.03 | 0 | — |