Klaus is right.
LODs for individual meshes allow you to have set quality levels for decimation of the mesh so that it’s less triangle heavy. So if you open the Static Mesh editor you can import manually separate LODs, use a decimation tool like Simplygon in the editor, or have these already included with your import of your FBX from your modeling application.
HLODs allow you to clump actors together based on values entered into the HLOD Outliner’s LODs. The cluster generation settings allow you to specify how the bounds for these generate clusters are formed throughout the level. After you generate clusters and then finalize with the Generate Proxy Meshes you will end up with something like this:
This requires no setup in the original Static Mesh for LODs. You get static meshes that are individual actors initially like here:
After you’ve generated the HLODs and you back the camera away to the specified screen size you get the switch to the HLOD, which is just the combined actors, in this case the two chairs and table.
This HLOD actor can be opened on it’s own in the static mesh editor. It’s now it’s own asset that is tied to the HLOD generation settings to know when it should transition for the other actors. This is a great way to reduce draw calls and regain performance in dense environments. HLODs just make use of clumping actors together at different distances and in that regard is separate than mesh reduction/decimation techniques.
In 4.13 and previous versions HLODs were not decimated, which means they could still be vert heavy unless you had a license for Simplygon. In 4.14 Epic has made their own decimation tool for mesh reduction, so you can open the HLOD asset and setup separate LODs to reduce the vert count.
Performance wise, you should make use of both HLOD and LODs, especially for densely packed environments where you need to save on performance by reducing drawcalls.