Not that I have anything against Brython (which imo is a cool project, although I'm not so sure if it's practical yet), but since you asked for it, here's an example where it does differ from python:

  x = 0
  
  def impure():
      global x
      x += 1
      return x

  print(0 < impure() <= 1)

In python, impure() would only be evaluated once returning True, but Brython's chaining evaluates it twice, and so it returns False instead. Of course, this example isn't exactly something you'd usually see in actual code (heck, it's probably even bad practice in most situations) BUT the point here is that while most of python might be compatible with this, there exists some inconsistencies that causes one's previously working python code to fail (and these bugs would probably be very difficult to trace). Of course, Brython is still young and has time to hopefully fix these issues. (In fact, this is the second time I've tried this project. Back then they didn't even have the chaining feature working, so it's nice that they are making progress.)

There's no support for metaclasses. type is missing entirely. nonlocal doesn't work. docstrings and some other double underscore names are just missing (__class__, __name__, etc.). Unicode literals don't work. dis obviously won't work since there isn't a bytecode compilation step. Scoping, like in the following code, behaves improperly:

  x = 1  
  def test():  
      print(x)  
      x = 2

That's not right, it is a translator/compiler. It simply translates a python code string into a JS code string, and then calls `eval()` on the JS code string. The `eval()` step can optionally be skipped so that you can precompile. This is much easier to implement than an interpreter anyway.

Yield support is buggy:

    def f():
        n = 0
        while n < 10:
          yield n
          n += 1
          print(n)
    g = f()
    print(next(g))

This prints 0, 1, ... 10, 0 instead of just 0