List your company

Multimodal Mathematical Reasoning on MathVista (MathVista score, Avg-R)

76.4MathVista ScoreSPOT-E
57.26462.23267.272.168May 12, 2026May 18, 2026May 24, 2026May 30, 2026Jun 5, 2026Jun 11, 2026Jun 18, 2026

Evaluation results

MethodLinks
SPOT-EBase Model=Qwen3-VL-32B2026.0676.4—
SPOT-EBase Model=Qwen2.5-VL-32B2026.0675.3—
Qwen3-VL-32BBase Model=Qwen3-VL-32B2026.0675.1—
SPOT-EBase Model=InternVL2.5-26B2026.0674.3—
Qwen2.5-VL-32BBase Model=Qwen2.5-VL-32B2026.0673.8—
InternVL2.5-26BBase Model=InternVL2.5-26B2026.0672.5—
GAP (LH+PCA+DA)Latent Head (LH)=true, Difficulty-Aware supervision (DA)=true, PCA alignment (PCA)=true2026.057153.97
GAP (LH+PCA)Latent Head (LH)=true, PCA alignment (PCA)=true2026.0570.252.48
GAP (LH+DA, no PCA)Latent Head (LH)=true, Difficulty-Aware supervision (DA)=true, PCA alignment (PCA)=false2026.0568.950.05
GAP (LH)Latent Head (LH)=true2026.0568.649.15
Qwen2.5-VL 7BParameters=7B2026.0568.552.62
Dense Cap SFT2026.0567.747.24
LVR2026.056747.66
Monet-7BParameters=7B2026.0561.347.99
SPOT-EBase Model=LLaVA-OV-72B2026.0660.3—
LLaVA-OV-72BBase Model=LLaVA-OV-72B2026.0658—