List your company

Visual Question Answering on GQA (Accuracy, Cost)

75.77AccuracyOriginal
27.09839.73452.3765.006Feb 3, 2026

Evaluation results

MethodLinks
OriginalModel=LLaVA-1.5-7B-hf2026.0275.77120
OriginalModel=Qwen2-VL-72B-Instruct2026.0274.35120
JPEG Q10Model=LLaVA-1.5-7B-hf2026.0274.1163.9
OriginalModel=Pixtral-12B-24092026.0273.52120
OriginalModel=Qwen3-VL-235B-A22B-Instruct2026.0272.98120
JPEG Q10Model=Qwen2-VL-72B-Instruct2026.0271.5763.9
VOIModel=LLaVA-1.5-7B-hf2026.0271.1826.86
VOIModel=Qwen2-VL-72B-Instruct2026.0271.0345.37
JPEG Q10Model=Qwen3-VL-235B-A22B-Instruct2026.0270.463.9
JPEG Q10Model=Pixtral-12B-24092026.0269.8163.9
VOIModel=Qwen3-VL-235B-A22B-Instruct2026.0268.2844.7
VOIModel=Pixtral-12B-24092026.0267.3745.66
Resize 32x32Model=LLaVA-1.5-7B-hf2026.0267.3718.1
OriginalModel=Qwen2.5-VL-7B-Instruct2026.0267.22120
CaptionModel=LLaVA-1.5-7B-hf2026.0264.099.1
OriginalModel=Llama-4-Maverick (17Bx12E)2026.0263.56120
JPEG Q1Model=LLaVA-1.5-7B-hf2026.0263.5145.4
JPEG Q10Model=Llama-4-Maverick (17Bx12E)2026.0263.1763.9
VOIModel=Qwen2.5-VL-7B-Instruct2026.0262.9259.66
JPEG Q1Model=Qwen2-VL-72B-Instruct2026.0262.6845.4
JPEG Q1Model=Qwen3-VL-235B-A22B-Instruct2026.0262.5845.4
VOIModel=Llama-4-Maverick (17Bx12E)2026.0260.8741.64
JPEG Q1Model=Pixtral-12B-24092026.0260.5345.4
JPEG Q10Model=Qwen2.5-VL-7B-Instruct2026.0260.1463.9
Resize 32x32Model=Qwen3-VL-235B-A22B-Instruct2026.0254.0818.1
JPEG Q1Model=Llama-4-Maverick (17Bx12E)2026.0252.7645.4
Resize 32x32Model=Qwen2-VL-72B-Instruct2026.0249.6318.1
JPEG Q1Model=Qwen2.5-VL-7B-Instruct2026.0248.6645.4
Resize 32x32Model=Qwen2.5-VL-7B-Instruct2026.0246.718.1
Resize 32x32Model=Llama-4-Maverick (17Bx12E)2026.0243.9218.1
CaptionModel=Pixtral-12B-24092026.0242.459.1
CaptionModel=Qwen2.5-VL-7B-Instruct2026.0241.289.1
Resize 32x32Model=Pixtral-12B-24092026.0238.118.1
CaptionModel=Llama-4-Maverick (17Bx12E)2026.0234.939.1
CaptionModel=Qwen2-VL-72B-Instruct2026.0232.349.1
CaptionModel=Qwen3-VL-235B-A22B-Instruct2026.0228.979.1