Level-of-Detail (LOD) and culling are widely used optimization methods in rasterization rendering. However, in ray tracing, conventional culling methods often fail due to global illumination effects such as reflection, where all geometry is potentially visible. This makes it challenging to identify and simplify redundant geometry without compromising visual accuracy. In this work, we are exploring a novel LOD method for ray-traced scenes that dynamically adjusts the level of detail based on a geometry's estimated contribution to the rendered frame. The system gathers information during the ray tracing process, estimates the contribution based on distance and material information, and selects a dynamic level of detail for the geometry. We hope to make an approach that balances performance and visual fidelity, providing a robust solution to this long-standing challenge.