List your company

Multimodal Reasoning on MMBench CN

82AccuracyInstruct
50.69658.82366.9575.077Aug 8, 2024Nov 30, 2024Mar 24, 2025Jul 16, 2025Nov 7, 2025Mar 1, 2026Jun 23, 2026

Evaluation results

MethodLinks
InstructModel Scale=14B, Base Model=InternVL32026.0382
IPOModel Scale=14B, Base Model=InternVL32026.0382
IPOModel Scale=8B, Base Model=InternVL32026.0382
β-DPOModel Scale=14B, Base Model=InternVL32026.0382
VanillaToken Budget=1296 Tokens (Upper Bound), Backbone=Qwen-2.5-VL-7B2026.0281.8
DPOModel Scale=14B, Base Model=InternVL32026.0381.6
InternVL2-8B-FTRP fine-tuning=None2024.0881.5
ACPOModel Scale=8B, Base Model=InternVL32026.0381.5
InternVL2-8B + RPRP fine-tuning=Object Replacement Data2024.0881.4
InstructModel Scale=8B, Base Model=InternVL32026.0381.2
DPO-ShiftModel Scale=14B, Base Model=InternVL32026.0381
ACPOModel Scale=14B, Base Model=InternVL32026.0381
PIO-FVLMToken Budget=Retain 33.30% Tokens, Venue=Ours, Backbone=Qwen-2.5-VL-7B2026.0280.5
DARTToken Budget=Retain 33.30% Tokens, Venue=EMNLP'25, Backbone=Qwen-2.5-VL-7B2026.0279
SimPOModel Scale=14B, Base Model=InternVL32026.0379
PIO-FVLMToken Budget=Retain 22.2% Tokens, Venue=Ours, Backbone=Qwen-2.5-VL-7B2026.0278.1
DPOModel Scale=8B, Base Model=InternVL32026.0378
DPO-ShiftModel Scale=8B, Base Model=InternVL32026.0378
DARTToken Budget=Retain 22.2% Tokens, Venue=EMNLP'25, Backbone=Qwen-2.5-VL-7B2026.0276.5
Qwen3-VL-2BRatio=100%2026.0575.93
F3ARatio=40%2026.0575.58
DivPruneRatio=60%2026.0575.51
PIO-FVLMToken Budget=Retain 11.1% Tokens, Venue=Ours, Backbone=Qwen-2.5-VL-7B2026.0275.5
F3ARatio=60%2026.0575.49
VisionZipRatio=60%2026.0575.44
CDPrunerRatio=60%2026.0575.33
SimPOModel Scale=8B, Base Model=InternVL32026.0375
CDPrunerRatio=40%2026.0574.79
β-DPOModel Scale=8B, Base Model=InternVL32026.0374.5
VisionZipRatio=40%2026.0574.33
FastVRatio=60%2026.0574.19
DivPruneRatio=40%2026.0573.94
F3ARatio=20%2026.0573.87
FastVRatio=40%2026.0573.2
CDPrunerRatio=20%2026.0572.41
DARTToken Budget=Retain 11.1% Tokens, Venue=EMNLP'25, Backbone=Qwen-2.5-VL-7B2026.0271.5
DivPruneRatio=20%2026.0571.19
VisionZipRatio=20%2026.0570.87
FastVRatio=20%2026.0570.71
LLaVA-1.5-13B + MMFuserLLM=Vicuna-13B2024.1063.8
LLaVA-1.5-13BLLM=Vicuna-13B2024.1063.6
LLaVABackbone=LLaVA-NeXT-7B, Tokens Retained=2880, Retention Ratio=100%2025.0860.6
VanillaVisual Token Budget=2880, Token Retention Ratio=100%2026.0660.6
LLaVA-1.5-7B + MMFuserLLM=Vicuna-7B2024.1060.1
LLaVA-1.5-7B + RPRP fine-tuning=Object Replacement Data2024.0859.8
MGM-7B + RPRP fine-tuning=Object Replacement Data2024.0859.6
HiPruneBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.0859.3
HiPrune++Backbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.0859.1
MGM-7BRP fine-tuning=None2024.0858.9
SparseVLMBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.0858.6
LLaVA-1.5-7BLLM=Vicuna-7B2024.1058.3
LLaVA-1.5-7BRP fine-tuning=None2024.0858.3
VanillaBackbone=LLaVA-1.5-7B, Retained Tokens=576, Compression Ratio=100%2026.0458.1
VisionZipBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.0858.1
FastVBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.0457.5
DARTBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%2026.0457.5
VisionZipBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.0457.4
HiPrune++Backbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.0857.4
EvoCompBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%, Number of compressor layers (l)=22026.0457.2
EvoCompBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%, Number of compressor layers (l)=22026.0457.1
DARTBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.0457
HiREDBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.0857
HiPruneBackbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.0857
VisionZipBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%2026.0456.9
PyramidDropBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.0456.8
Qwen-VL-ChatLLM=Qwen-7B2024.1056.7
SparseVLMBackbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.0856.7
FastVBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%2026.0456.6
PyramidDropBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%2026.0456.6
DivPruneBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.0856.4
HiREDBackbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.0856.4
PDropVisual Token Budget=320, Token Retention Ratio=11.1%2026.0656.2
EvoCompBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%, Number of compressor layers (l)=02026.0456.1
EvoCompBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%, Number of compressor layers (l)=02026.0456.1
VisPrunerBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.0856
MustDropBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.0455.8
TRIMBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.0855.8
VisionZipBackbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.0855.6
VisPrunerBackbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.0855.4
VisionZipBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%2026.0455.3
PruMergeVisual Token Budget=320, Token Retention Ratio=11.1%2026.0655.3
MustDropBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%2026.0455.2
PDropBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.0855.2
DivPruneBackbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.0855.2
EvoCompBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%, Number of compressor layers (l)=02026.0455.1
HiREDBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.0454.7
SparseVLMVisual Token Budget=320, Token Retention Ratio=11.1%2026.0654.5
PriorTRVisual Token Budget=320, Token Retention Ratio=11.1%2026.0654.4
SparseVLMBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.0453.7
EvoCompBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%, Number of compressor layers (l)=22026.0453.7
DARTBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%, Number of compressor layers (l)=02026.0453.6
HiREDBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%2026.0453.6
FastVBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.0853.5
DARTBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%2026.0453.2
MustDropBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%2026.0453.1
PruMergeBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.0452.9
FastVBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%2026.0452.8
DivPruneBackbone=LLaVA-NeXT-7B, Tokens Retained=160, Retention Ratio=5.6%2025.0852.3
DARTBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%, Number of compressor layers (l)=02026.0451.9
FastVVisual Token Budget=320, Token Retention Ratio=11.1%2026.0651.9