List your company
ResearchFormerly WizWand

Interleaved Image Multimodal Understanding on BLINK

66.3ScoreInternVL3-78B
-2.65215.24933.1551.051Apr 14, 2025Jun 5, 2025Jul 27, 2025Sep 17, 2025Nov 8, 2025Dec 30, 2025Feb 21, 2026
Updated 6mo ago

Evaluation results

MethodLinks
InternVL3-78Bnumber of parameters=78B2026.0266.3
InternVL3-38Bnumber of parameters=38B2026.0264.49
Qwen2.5-VL-32Bnumber of parameters=32B2026.0259.34
TPRU-32Bnumber of parameters=32B2026.0258.81
TPRU-7Bnumber of parameters=7B2026.0255.86
Qwen2.5-VL-7Bnumber of parameters=7B2026.0254.76
Qwen2.5-VL-3Bnumber of parameters=3B2026.0248.97
Full ModelSparsity Ratio=0%, Base Model=LLaVA-OneVision2025.0448.4
TPRU-3Bnumber of parameters=3B2026.0248.13
SparseGPTSparsity Ratio=60%, Base Model=LLaVA-OneVision2025.0446.2
TAMPSparsity Ratio=60%, Base Model=LLaVA-OneVision2025.0445.9
ECOFLaPSparsity Ratio=60%, Base Model=LLaVA-OneVision2025.0445.2
WandaSparsity Ratio=60%, Base Model=LLaVA-OneVision2025.0444
OWLSparsity Ratio=60%, Base Model=LLaVA-OneVision2025.0443.5
MRoPE-IBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=MRoPE-I2025.1037.88
CircleRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=CircleRoPE2025.1037.52
MHRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=MHRoPE2025.1037.22
MRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=MRoPE2025.1036.8
Vanilla RoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=Vanilla RoPE2025.1036.33
HoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=HoPE2025.1034.46
VideoRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=VideoRoPE2025.1034.44
MagnitudeSparsity Ratio=60%, Base Model=LLaVA-OneVision2025.040