How much IP address needs to be cut to effectively anonymise HTTP logs?

I'd like to open the data behind research on https://www.earth.org.uk/RSS-efficiency.html and internally I am retaining RSS feed fetch logs with the last IPv4 octet hidden, but what is a reasonable level to hide to avoid re-identification (if I remove other obvious User-Agent IDs)? The last two octets? Seeing some of the IP helps decide if a visitor is a repeat, or emerging from a cloud with a randomish IP, etc.