Not much; just, oh, stuff like floating-point math and trig functions; integer math, associative arrays, access to environment variables, named functions with arguments and full recursion; control and looping structures like if, for and while, file I/O and redirection, string literals with C escapes, ...
$ awk '$0=sin($1) + cos($2)'
0 1
0.540302
1 0
1.84147
How about sed?Just kidding ...
The one huge thing that sed is particularly bad at (unless I don't know the particular trick yet) is to split fields on a pattern. I frequently have pipes where sed is used for most of the text-processing, but then cut is used to select one (or multiple) character-separated fields, or awk is used to select one (or multiple) whitespace-separated fields. I only go one step further (usually to Perl) when operations involve multiple lines, e.g. when columns need to be summed up.
Wrote this earlier in this same thread: