If you always have some mesh a ray can collide with (like ground, which you seem to have):
- Create a collision box (Cube or Box3 with size that will determine your selection radius.)
- Ray-cast mouse location on the scene and move the collision box to where the collision happens.
- Check which models’ bounding boxes collide with the collision box.
- For each colliding model detect which faces overlap with the collision box (intersectsTriangle may also help), ignore faces which point in the opposite direction.
Otherwise, if you don’t have a ground mesh, and I understood correctly, you can’t really do what you are trying to do, since cursor has no depth - some faces that are visually next to the cursor in 2D can be far apart in 3D. One more thing you could do it cast more rays with some offset - create a Vector2 with (mouse.x + offset , mouse.y), duplicate it 8 times and rotate around the origin by 45°. Additional rays will then pick up objects that are nearby the mouse position, with some precision. If performance gets hit - you can do the collision testing on an offscreen canvas worker and return only the collisions to the main thread.