Why does RAM usage differ between /api/monitor/metrics (Prometheus) and Zabbix?

I built a monitoring system in Grafana using the IRIS API /api/monitor/metrics (reading with Prometheus) but I noticed that the RAM usage shown was below that shown by the operating system.
I installed the Zabbix agent and the usage values ​​were higher, but with a line with the same highs and lows but shifted.
The metric in the API is iris_phys_mem_percent_used, described as "Percent of physical memory (RAM) currently in use", in Zabbix it is the Item tag: "component: memory" item: "Memory utilization".

Image below of the graphs in Grafana:

Using the command "free -h" the server returns:

            total        used        free         shared     buff/cache   available
Mem:  15Gi       8.1Gi        6.8Gi       3.2Gi        3.7Gi              7.0Gi
 

What exactly does the iris_phys_mem_percent_used metric return?
Would it be the use of physical memory exclusively by IRIS? If so, is there any other metric in the IRIS API that returns the total RAM usage of the server?