@ThingumaBob mentioned that it is the projection used in many popular web mapping applications (Google/Bing/OpenStreetMap/etc). Sometimes also known as EPSG:900913.
EPSG 4326 uses a coordinate system the same as a Globe (curved surface) but EPSG 3857 uses a coordinate system the same as a MAP (flat surface).
ESPG 4326(WGS 84 ):The World Geodetic System of 1984 is the geographic coordinate system (the three-dimensional one) used by GPS to express locations on the earth. WGS84 is the defined coordinate system for GeoJSON, as longitude and latitude in decimal degrees. For the most part, when you describe a lon/lat coordinate location, it’s based on the EPSG:4326 coordinate system. This is how we store data at Mapbox.
There is no way to visualize the WGS84 coordinate system on a two-dimensional plane (map), so most software programs project these coordinates using an equirectangular projection (Plate-Carrée) (Esri uses EPSG:54001).
EPSG:3857 (Pseudo-Mercator): The projected Pseudo-Mercator coordinate system takes the WGS84 coordinate system and projects it onto a square. (This projection is also called Spherical Mercator or Web Mercator.) But not all of it – the bounds of Pseudo-Mercator are limited to approximately 85.06º North and South latitude. This projection was first introduced by Google and is used in almost 100% of web maps, but it’s a strange one: the projection uses the WGS84 coordinate system, which uses the WGS84 ellipsoid, but projects the coordinates onto a sphere.
This method results in a square-shaped map (which computers really want) but there is no way to programmatically represent a coordinate system that relies on two different ellipsoids, which means software programs have to improvise. And when software programs improvise, there is no way to know if the coordinates are consistent across programs. This makes EPSG:3857 great for visualizing on computers but not reliable for data storage or analysis.
For the most part, web maps rely on data stored with WGS84 coordinates (in some programs this is called “unprojected” data) and then visualize the data using Pseudo-Mercator.