User tests AI models via playable 3D games for coherence

A Reddit user evaluated Fable 5, Opus 4.8, GPT-5.6, and other models through custom 3D games to assess spatial, temporal, and causal reasoning. The tests focus on consistency in simulated environments as a proxy for world-model capabilities.