Comment by DarkPlayer

We observed the same behavior. Each request used a different IP address and a random user agent. In our case, most of the IP addresses belonged to Chinese ISPs. They went to great lengths to avoid being blocked, but at the same time used user agents such as Windows 95/98 or IE 5. Fortunately, the combination of the odd user agents and the fact that they still use HTTP/1.1 makes them somewhat easy to identify. So you can use a captcha on more expensive endpoints to block them.