ResearchBenchmarksMultimodal Reasoning on AI2D (Accuracy)Follow62.53AccuracyESC50.77853.82956.8859.931Jul 1, 2026Evaluation ResultsMethodMethodLinksAccuracyESCModel=Qwen2 [92]Model=Qwen2 [92]2026.0762.53BaselineModel=Qwen2 [92]Model=Qwen2 [92]2026.0760.98ESCModel=LLaVA [50]Model=LLaVA [50]2026.0753.72BaselineModel=LLaVA [50]Model=LLaVA [50]2026.0751.23