As Jon describes "They have an arp table for the same reason as routers, switches etc."
Some additional information, which you may, or may not, find helpful . . .
The reason for an ARP table/cache is to avoid the need to ARP for a destination IP's MAC. Generally, ARP table/cache entries are only retained for some period of time (often configurable). When the ARP entry expires, the host will once again ARP and load an ARP entry into the ARP table/cache. (If a destination IP's MAC changes, a host will send the frame to the wrong MAC unless it somehow "knows" of the change. When this happens, you need to flush the ARP cache.) (BTW, it's often also possible to sometimes load an ARP table/cache with manual entries, which do not time out.)
Jon also notes a PC/Server host usually only has ARP entries for one subnet and a gateway, while L3 devices have entries for multiple networks. That's true, it's generally as simple as the host has ARP entries for all IPs it has been {recently) sending traffic to. As L3 devices often have interfaces in multiple networks, the ARP cache reflects that. A PC or server host, if also having multiple network connected interfaces (i.e. multi-homed), ARP cache may also appear more like a L3 device.
