List your company

Diagram Understanding on AI2D (Efficiency Metrics)

87.14AI2D ScoreBaseline
64.842470.631276.4282.2088Nov 22, 2025Dec 19, 2025Jan 16, 2026Feb 13, 2026Mar 12, 2026Apr 9, 2026May 7, 2026

Evaluation results

MethodLinks
BaselineVision Token Retention Ratio=100%, Saving FLOPs. (%)=02025.1187.1479.160—
FastMMoE†Vision Token Retention Ratio=75%, lv=6, Kv=5, Saving FLOPs. (%)=37.592025.1186.3777.41.76—
FastMMoE*Vision Token Retention Ratio=75%, Saving FLOPs. (%)=21.992025.1186.2777.561.6—
FastVVision Token Retention Ratio=75%, Saving FLOPs. (%)=21.992025.1185.5976.932.22—
FastMMoE†Vision Token Retention Ratio=75%, lv=3, Kv=4, Saving FLOPs. (%)=44.652025.1185.4676.842.31—
FastMMoE*Vision Token Retention Ratio=50%, Saving FLOPs. (%)=44.232025.1184.4976.352.8—
InternVL3Params=8B2026.0483.9———
SparseVLMVision Token Retention Ratio=75%, Saving FLOPs. (%)=21.992025.1183.8170.069.09—
FastMMoE†Vision Token Retention Ratio=50%, lv=6, Kv=5, Saving FLOPs. (%)=55.022025.1183.7175.633.53—
FastMMoE†Vision Token Retention Ratio=50%, lv=3, Kv=3, Saving FLOPs. (%)=64.552025.1183.4874.694.47—
Qwen3-VLParams=8B2026.0483.2———
Qwen2.5-VLParams=7B2026.0482.5———
FastVVision Token Retention Ratio=50%, Saving FLOPs. (%)=44.232025.1182.4274.334.82—
BaselineRetention Ratio (Vision Tokens)=100%2025.1182.3872.6800
InternVL3.5Params=8B2026.0482.3———
FastMMoE*Vision Token Retention Ratio=25%, Saving FLOPs. (%)=64.302025.1182.1273.525.63—
FastMMoE*Retention Ratio (Vision Tokens)=75%2025.1182.0672.490.1921.09
FastVRetention Ratio (Vision Tokens)=75%2025.1181.9972.180.5—
SparseVLMRetention Ratio (Vision Tokens)=75%2025.1181.5771.720.96—
FastMMoE†Retention Ratio (Vision Tokens)=75%, lv=10, Kv=22025.1181.54720.6839.15
FastMMoE†Vision Token Retention Ratio=25%, lv=6, Kv=5, Saving FLOPs. (%)=70.472025.1181.3871.897.27—
FastMMoE*Retention Ratio (Vision Tokens)=50%2025.1181.3571.940.7443.07
FastVRetention Ratio (Vision Tokens)=50%2025.1181.1971.910.77—
SparseVLMRetention Ratio (Vision Tokens)=50%2025.1180.5170.312.37—
FastMMoE†Retention Ratio (Vision Tokens)=50%, lv=2, Kv=22025.1180.3870.572.1161.76
MUSE-3BType=Unified, # Params=2B+1.6B2026.0580.2———
FastMMoE*Retention Ratio (Vision Tokens)=25%2025.1179.1570.322.3663.66
SparseVLMRetention Ratio (Vision Tokens)=25%2025.1179.0266.616.07—
FastVRetention Ratio (Vision Tokens)=25%2025.1178.8270.022.65—
Show-o2Type=Unified, # Params=7B2026.0578.6———
UniLIP-3BType=Unified, # Params=2B+1.6B2026.0578.6———
InternVL3Type=Und. Only, # Params=1.8B2026.0578.5———
SparseVLMVision Token Retention Ratio=50%, Saving FLOPs. (%)=44.232025.1177.3662.5516.6—
FastVVision Token Retention Ratio=25%, Saving FLOPs. (%)=64.302025.1177.1469.219.95—
UniDriveVLAParams=8B2026.0476.3———
MUSE-1BType=Unified, # Params=1B+0.6B2026.0572.5———
SparseVLMVision Token Retention Ratio=25%, Saving FLOPs. (%)=64.302025.1172.1254.7324.43—
UniLIP-1BType=Unified, # Params=1B+0.6B2026.0570.7———
GPT-5 Nano2026.0465.7———