ResearchBenchmarksSpatial Reasoning on OMNI3D-BENCH 100 held out queriesSpatial Reasoning on OMNI3D-BENCH 100 held out queriesFollow44AccuracyVALOR16.75223.82630.937.974Dec 9, 2025Evaluation resultsMethodMethodLinksAccuracyVALORBase Model=Qwen3-8BBase Model=Qwen3-8B2025.1244VALOR-RLBase Model=Qwen3-8BBase Model=Qwen3-8B2025.1243.9VADARBase Model=GPT-4oBase Model=GPT-4o2025.1238.9ViperGPTBase Model=GPT-3.5Base Model=GPT-3.52025.1225.5VisProgBase Model=GPT-4oBase Model=GPT-4o2025.1217.8