List your company

Natural Language Visual Reasoning on NLVR2

88.8AccuracyGPT-4V
5.49627.12348.7570.377Sep 14, 2023Feb 19, 2024Jul 26, 2024Dec 31, 2024Jun 7, 2025Nov 12, 2025Apr 20, 2026

Evaluation results

MethodLinks
GPT-4VParameter=-2026.0488.8—
Ours (masked) (LLaVA-OV-7B)Model Size=7B, Backbone=LLaVA-OV, Attention Masking=True2026.0187.3—
Idefics2Parameter=8B2026.0486.9—
LLaVA-OV-7BModel Size=7B2026.0184.2—
Qwen2.5-VL + S2H-DPOParameter=7B2026.0474.67—
Qwen2.5-VLParameter=7B2026.0474.28—
Qwen2.5-VL + MIA-DPOParameter=7B2026.0474.18—
Ours (LLaVA-OV-1.5B)Model Size=1.5B, Backbone=LLaVA-OV, Attention Masking=False2026.0173.7—
EWCCheckpoint=Task 2: NLVR2, Backbone=ViLT2026.0372.85—
EWC-DRCheckpoint=Task 2: NLVR2, Backbone=ViLT2026.0372.77—
LLaVA-OV-1.5BModel Size=1.5B2026.0170.9—
EWC-DRCheckpoint=Task 3: SNLI-VE, Backbone=ViLT2026.0370.659.31
Ours (masked) (LLaVA-OV-1.5B)Model Size=1.5B, Backbone=LLaVA-OV, Attention Masking=True2026.0169—
Ours (LLaVA-OV-0.5B)Model Size=0.5B, Backbone=LLaVA-OV, Attention Masking=False2026.0168—
MMICL2023.0966.6—
EWC-DRCheckpoint=Task 4: VCR, Backbone=ViLT2026.0366.5127.48
Ours (masked) (LLaVA-OV-0.5B)Model Size=0.5B, Backbone=LLaVA-OV, Attention Masking=True2026.0165.1—
EWCCheckpoint=Task 3: SNLI-VE, Backbone=ViLT2026.0364.9534.59
LLaVA-OV-0.5BModel Size=0.5B2026.0161.2—
InstructBLIPParameter=13B2026.0460.3—
LLaVA-v1.6Parameter=7B2026.0458.9—
Qwen-VL-ChatParameter=7B2026.0458.7—
CogVLMParameter=17B2026.0458.6—
Emu2-ChatParameter=37B2026.0458.2—
VideoLLaVAParameter=7B2026.0456.5—
LLaVA-v1.5 + S2H-DPOParameter=7B2026.0455.59—
LLaVA-v1.5 + MIA-DPOParameter=7B2026.0454.2—
InstructionBlip2023.0953.95—
EWCCheckpoint=Task 4: VCR, Backbone=ViLT2026.0352.1390.66
LLaVA-v1.5Parameter=7B2026.0452.1—
LLaVA-v1.5 + LLaVA-RLHFParameter=7B2026.0451.8—
LLaVA-v1.5 + HA-DPOParameter=7B2026.0451.6—
FuyuParameter=8B2026.0451.1—
Qwen3-VL + S2H-DPOParameter=2B2026.0450.61—
Qwen3-VLParameter=2B2026.0449.71—
OTTER2023.0947.2—
Qwen2VL-2BModel Size=2B2026.0141.6—
Qwen2VL-7BModel Size=7B2026.0141.5—
LLaVA-v1.5 + POVIDParameter=7B2026.0421.4—
InternVL2-2BModel Size=2B2026.0118.9—
InternVL2-8BModel Size=8B2026.018.7—