The documentation for the EXTENDED_ARG instruction in the dis module documentation refers to the way the opcode worked before 3.6: https://docs.python.org/3.6/library/dis.html#opcode-EXTENDED_ARG

As I understand, since moving to 2-byte wordcode in 3.6, each EXTENDED_ARG effectively adds a byte to the argument of the next instruction and they can be chained to allow up to a 32-bit argument. The current documentation refers the 2-byte arguments from the older bytecode used in 3.5 and below.

I'm trying to think of a clear and concise wording for how it works now and will add a PR to fix this issue unless someone gets to it before me.