List your company

Visual Question Answering on CVQA (Global, SEA, Others)

67.59Global Accuracy (CVQA)SEA-Gemma-3 10%
22.651634.318345.98557.6517Apr 13, 2026

Evaluation results

MethodLinks
SEA-Gemma-3 10%Number of Parameters=27B2026.0467.5969.1267.0666.2670.9471.4375.1474.363.6467.7363.562.3975.8745.6864.4265.0281.094765.9553.6953.1977.9377.2350.6469.5747.280.6376.4981.565.7869.9280.9574.575.69—
Gemma-3Number of Parameters=27B2026.0465.6966.5865.3966.566.0167.6272.6673.3662.9664.545962.8272.0342.9662.2662.0780.643.564.1153.253.6276.974.7550.3266.8945.7980.6375.1780.565.3369.3476.1971.573.39—
MolmoNumber of Parameters=72B2026.0460.8662.6260.2563.8364.5369.8479.2958.4154.8858.175245.368.8835.0657.6858.6278.1141.557.0657.1440.4374.1468.8147.7669.942.0677.4670.28445.7871.5567.775769.5—
Aya-VisionNumber of Parameters=32B2026.0460.1864.4558.767.8866.3472.3877.4861.6855.5661.7552.5329.1864.3139.3656.4968.4778.113856.1359.1140.438066.1736.0166.2236.4578.0174.178039.5673.2566.1857.7966.69—
Qwen-2.5-VLNumber of Parameters=72B2026.0459.0862.1758.0166.565.0262.580.9858.8855.2251.7956.536.4861.2737.7861.9961.0875.1236.5557.9858.6238.377.5961.3939.168.5635.0576.7675.837945.571.4455.3158.564.92—
Qwen-3 VLNumber of Parameters=35B2026.0458.8363.1857.3266.2664.0470.1678.2861.6854.5554.9855.542.7464.3434.8154.4557.6475.6234.555.5258.6232.7777.2463.8644.2361.238.3276.0670.276.543.5668.260.076166.99—
Llama-3.2 VisionNumber of Parameters=11B2026.0451.5353.6350.8156.3151.7256.1962.858.4147.8151.794439.3255.5934.5749.0649.2668.164453.9950.7435.3259.6648.0239.4254.5234.1157.7558.9466.54857.0253.1155.555.98—
PixtralNumber of Parameters=12B2026.0449.0157.7145.9962.8664.5361.968.9351.8751.1851.394932.9148.2535.822.9143.3530.8541.557.6749.2634.4773.4531.1939.7464.2135.5173.5967.883728.4469.1360.0732.548.02—
SEA-Gemma-3Number of Parameters=27B2026.0431.9426.9533.6821.624.1429.8440.7340.6522.2223.912.526.0733.5727.6534.2332.0242.7925.531.2929.0628.5142.7634.1623.438.824.7741.240.7345.529.7837.6136.263236.93—
SEA-SigLIP2Merge weight (beta)=50%, Merge strategy=linear merge, Evaluation protocol=Zero-Shot2026.0427.5225.527.97————————————————————————————————
SEA-SigLIP2Merge weight (beta)=75%, Merge strategy=linear merge, Evaluation protocol=Zero-Shot2026.0427.1227.5127.03————————————————————————————————
SEA-SigLIP2 (w/o merge)Merge strategy=w/o merge, Evaluation protocol=Zero-Shot2026.0426.7525.2927.07————————————————————————————————
Google SigLIP2Evaluation protocol=Zero-Shot2026.0425.5124.0225.84————————————————————————————————
SEA-SigLIP2Merge weight (beta)=25%, Merge strategy=linear merge, Evaluation protocol=Zero-Shot2026.0424.3824.4424.36————————————————————————————————
BreadcrumbsModel=Breadcrumbs2026.05——————————————————————————————————34.24
DAREModel=DARE2026.05——————————————————————————————————38.42
DiM3-7BModel=DiM3-7B2026.05——————————————————————————————————43.78
LLaVA-v1.5-7BModel=LLaVA-v1.5-7B2026.05——————————————————————————————————39.32
NeuroMergingModel=NeuroMerging2026.05——————————————————————————————————39.38
Palo-13BModel=Palo-13B2026.05——————————————————————————————————39.59
Palo-7BModel=Palo-7B2026.05——————————————————————————————————37
PCB-MergingModel=PCB-Merging2026.05——————————————————————————————————7.8
STFModel=STF2026.05——————————————————————————————————34.59
Task ArithmeticModel=Task Arithmetic2026.05——————————————————————————————————40.47
TIES-MergingModel=TIES-Merging2026.05——————————————————————————————————39.22