The recent advancements in 3D modeling and depth estimation have seen significant innovations, particularly in the areas of autoregressive models and robust depth completion. Autoregressive models, traditionally used in language and image generation, are now being adapted for 3D shape modeling, offering more efficient and controllable generation processes. This shift is marked by the introduction of hierarchical and multi-scale approaches that reduce computational costs while maintaining geometric detail. Additionally, there is a growing focus on developing models that can generalize well across various datasets and scenarios, as evidenced by the advancements in depth completion techniques that incorporate multi-resolution integration and probability-based losses. These models are designed to handle sparse depth maps of varying densities, enhancing their applicability in real-world settings. Notably, the integration of large pre-trained models and novel training strategies is enabling high-quality 3D generation with versatile output formats and local editing capabilities. The field is also witnessing a trend towards domain-agnostic flow matching models, which simplify training processes and improve performance across different data modalities. Overall, the current research direction is towards more efficient, versatile, and robust 3D modeling and depth estimation techniques that can handle complex real-world scenarios.