List your company

Visual Question Answering on VQA v2 (Accuracy and Cost)

79.01AccuracyOriginal
31.991644.198356.40568.6117Feb 3, 2026

Evaluation results

MethodLinks
OriginalModel=Qwen3-VL-235B-A22B-Instruct2026.0279.01120
OriginalModel=Qwen2.5-VL-7B-Instruct2026.0277.98120
OriginalModel=Qwen2-VL-72B-Instruct2026.0277.16120
VOIModel=Qwen3-VL-235B-A22B-Instruct2026.0274.5752.04
JPEG Q10Model=Qwen2-VL-72B-Instruct2026.0274.2763.9
OriginalModel=Pixtral-12B-24092026.0274.06120
VOIModel=Qwen2-VL-72B-Instruct2026.0273.0148
OriginalModel=LLaVA-1.5-7B-hf2026.0272.8120
JPEG Q10Model=Qwen3-VL-235B-A22B-Instruct2026.0272.5863.9
JPEG Q10Model=Qwen2.5-VL-7B-Instruct2026.0272.163.9
VOIModel=Qwen2.5-VL-7B-Instruct2026.0272.0157.31
JPEG Q10Model=LLaVA-1.5-7B-hf2026.0270.2163.9
JPEG Q10Model=Pixtral-12B-24092026.0269.2163.9
VOIModel=Pixtral-12B-24092026.0268.6745.41
VOIModel=LLaVA-1.5-7B-hf2026.0268.0738.52
OriginalModel=Llama-4-Maverick (17Bx12E)2026.0267.37120
JPEG Q10Model=Llama-4-Maverick (17Bx12E)2026.0265.9963.9
VOIModel=Llama-4-Maverick (17Bx12E)2026.0264.6143.37
JPEG Q1Model=Qwen3-VL-235B-A22B-Instruct2026.0262.6245.4
JPEG Q1Model=Qwen2-VL-72B-Instruct2026.0260.4945.4
Resize 32x32Model=LLaVA-1.5-7B-hf2026.0258.9918.1
JPEG Q1Model=LLaVA-1.5-7B-hf2026.0258.2545.4
CaptionModel=LLaVA-1.5-7B-hf2026.0258.019.1
CaptionModel=Qwen3-VL-235B-A22B-Instruct2026.0257.369.1
JPEG Q1Model=Pixtral-12B-24092026.0257.2745.4
JPEG Q1Model=Qwen2.5-VL-7B-Instruct2026.0256.2645.4
JPEG Q1Model=Llama-4-Maverick (17Bx12E)2026.0250.4545.4
Resize 32x32Model=Qwen3-VL-235B-A22B-Instruct2026.0250.2118.1
CaptionModel=Pixtral-12B-24092026.0249.839.1
Resize 32x32Model=Qwen2.5-VL-7B-Instruct2026.0248.3818.1
CaptionModel=Llama-4-Maverick (17Bx12E)2026.0245.989.1
Resize 32x32Model=Qwen2-VL-72B-Instruct2026.0243.918.1
CaptionModel=Qwen2-VL-72B-Instruct2026.0240.219.1
Resize 32x32Model=Llama-4-Maverick (17Bx12E)2026.0239.4218.1
CaptionModel=Qwen2.5-VL-7B-Instruct2026.0236.659.1
Resize 32x32Model=Pixtral-12B-24092026.0233.818.1