List your company
ResearchFormerly WizWand

Multi-modal Understanding on MMBench (mean accuracy)

86.3Mean AccuracyOryx-1.5
41.68453.26764.8576.433Sep 19, 2024Jan 4, 2025Apr 21, 2025Aug 6, 2025Nov 21, 2025Mar 8, 2026Jun 23, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Oryx-1.5Size=32B2024.0986.3
LLaVA-OneVisionSize=72B2024.0985.6
OryxSize=34B2024.0984.5
Latent DenoisingArchitecture=Qwen-2.5-VL2026.0484.5
Qwen3-VL-4BModel=Qwen3-VL, Parameter Count=4B2026.0483.58
BaselineArchitecture=Qwen-2.5-VL2026.0483.1
VILA-1.5Size=40B2024.0982.4
Cambrian-1Size=34B2024.0981.4
OryxSize=7B2024.0981.4
Oryx-1.5Size=7B2024.0981.3
LLaVA-OneVision-7BModel=LLaVA-OneVision, Parameter Count=7B2026.0480.57
LLaVA-NeXTSize=34B2024.0979.3
InfiniteVL-4BModel=InfiniteVL, Parameter Count=4B, Result Source=InfiniteVL (Tao et al., 2025) paper2026.0479
SpB2.0-VL-5BModel=SpB2.0-VL, Parameter Count=5B2026.0478.39
Bunny-LLama3Size=8B2024.0977.2
Qwen2.5-VL-3BModel=Qwen2.5-VL, Parameter Count=3B2026.0476.88
Idefics2Size=8B2024.0976.7
Cambrian-1Size=8B2024.0975.9
VILA-1.5Size=8B2024.0975.3
InternVL2-4BModel=InternVL2, Parameter Count=4B, Result Source=OpenVLM Leaderboard (OpenCompass, 2025)2026.0473.6
Deepseek-VLSize=7B2024.0973.2
MonkeySize=7B2024.0972.4
LLaVA-NeXTSize=8B2024.0972.1
VITASize=8x7B2024.0971.8
MaTVLM-3BModel=MaTVLM, Parameter Count=3B, Result Source=InfiniteVL (Tao et al., 2025) paper2026.0469.4
Latent DenoisingArchitecture=LLaVA+SigLIP2026.0468.7
VanillaBackbone=LLaVA-1.5-13B, Retained Tokens=5762026.0668.5
PriorTRBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.0668.3
PDropBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.0667.5
SmolVLM2-2.2BModel=SmolVLM2, Parameter Count=2.2B2026.0467.48
SparseVLMBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.0667.4
PriorTRBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.0667.3
VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.0666.8
CAMDBase Model=LLaVA-1.52026.0366.6
VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.0666.5
FarSightBase Model=LLaVA-1.52026.0366
SparseVLMBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.0665.8
PriorTRBackbone=LLaVA-1.5-13B, Retained Tokens=642026.0665.8
Latent DenoisingArchitecture=LLaVA+CLIP2026.0465.7
PDropBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.0665
LLaVA-1.5-7BRetained Tokens=576, Retention Ratio=100%, Base Model=LLaVA-1.5-7B2025.0864.7
EVIANData Selection Strategy=10K subset2026.0464.63
CGDBase Model=LLaVA-1.5, Decoding Strategy=CGD2026.0364.5
BaselineArchitecture=LLaVA+SigLIP2026.0464.5
OPERABase Model=LLaVA-1.5, Decoding Strategy=OPERA2026.0364.4
LLaVA-1.5Decoding Strategy=Greedy2026.0364.3
BaselineArchitecture=LLaVA+CLIP2026.0464
VCDBase Model=LLaVA-1.5, Decoding Strategy=VCD2026.0363.9
HiPrune++Retained Tokens=192, Retention Ratio=33.3%, Base Model=LLaVA-1.5-7B2025.0863.5
VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=642026.0663.2
SCALEData Selection Strategy=10K subset2026.0463.18
BLIP-2Data Selection Strategy=10K subset2026.0463.17
ICDBase Model=LLaVA-1.5, Decoding Strategy=ICD2026.0363.1
CAMDBase Model=Video-LLaVA2026.0363.1
FarSightBase Model=Video-LLaVA2026.0362.8
HiPruneRetained Tokens=192, Retention Ratio=33.3%, Base Model=LLaVA-1.5-7B2025.0862.8
HiPrune++Retained Tokens=128, Retention Ratio=22.2%, Base Model=LLaVA-1.5-7B2025.0862.3
HiPruneRetained Tokens=128, Retention Ratio=22.2%, Base Model=LLaVA-1.5-7B2025.0862.2
BLIPData Selection Strategy=10K subset2026.0461.83
SparseVLMBackbone=LLaVA-1.5-13B, Retained Tokens=642026.0661.3
Video-LLaVADecoding Strategy=Greedy2026.0360.9
CAMDBase Model=Chat-UniVi2026.0360.6
HiPrune++Retained Tokens=64, Retention Ratio=11.1%, Base Model=LLaVA-1.5-7B2025.0860.3
ALBEFData Selection Strategy=10K subset2026.0460.03
FarSightBase Model=Chat-UniVi2026.0359.8
Full DataData Selection Strategy=Full 300K pool2026.0459.53
HiPruneRetained Tokens=64, Retention Ratio=11.1%, Base Model=LLaVA-1.5-7B2025.0859.5
PDropBackbone=LLaVA-1.5-13B, Retained Tokens=642026.0659.2
Qwen2.5-VLData Selection Strategy=10K subset2026.0457.96
FastVBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.0657.9
CLIPScoreData Selection Strategy=10K subset2026.0457.46
Chat-UniViDecoding Strategy=Greedy2026.0356.3
Cobra-3BModel=Cobra, Parameter Count=3B, Result Source=InfiniteVL (Tao et al., 2025) paper2026.0455.9
FastVBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.0654
RandomData Selection Strategy=10K subset2026.0453.53
FastVBackbone=LLaVA-1.5-13B, Retained Tokens=642026.0650.9
CAMDBase Model=InstructBLIP2026.0346.8
FarSightBase Model=InstructBLIP2026.0346.5
InstructBLIPDecoding Strategy=Greedy2026.0343.4