diff --git a/jedi/evaluate/context/typing.py b/jedi/evaluate/context/typing.py
index dbce7034..8a71ffcd 100644
--- a/jedi/evaluate/context/typing.py
+++ b/jedi/evaluate/context/typing.py
@@ -3,7 +3,7 @@ We need to somehow work with the typing objects. Since the typing objects are
 pretty bare we need to add all the Jedi customizations to make them work as
 contexts.
 """
-from jedi._compatibility import unicode
+from jedi._compatibility import unicode, force_unicode
 from jedi import debug
 from jedi.evaluate.cache import evaluator_method_cache
 from jedi.evaluate.compiled import builtin_from_name
@@ -376,6 +376,9 @@ class TypeVarClass(_BaseTypingContext):
             return None
         else:
             safe_value = method(default=None)
+            if self.evaluator.environment.version_info.major == 2:
+                if isinstance(safe_value, bytes):
+                    return force_unicode(safe_value)
             if isinstance(safe_value, (str, unicode)):
                 return safe_value
             return None
diff --git a/jedi/plugins/stdlib.py b/jedi/plugins/stdlib.py
index 3319046a..bd28bb05 100644
--- a/jedi/plugins/stdlib.py
+++ b/jedi/plugins/stdlib.py
@@ -403,13 +403,13 @@ def collections_namedtuple(obj, arguments):
     # Process arguments
     # TODO here we only use one of the types, we should use all.
     # TODO this is buggy, doesn't need to be a string
-    name = list(_follow_param(evaluator, arguments, 0))[0].get_safe_value()
+    name = force_unicode(list(_follow_param(evaluator, arguments, 0))[0].get_safe_value())
     _fields = list(_follow_param(evaluator, arguments, 1))[0]
     if isinstance(_fields, compiled.CompiledValue):
-        fields = _fields.get_safe_value().replace(',', ' ').split()
+        fields = force_unicode(_fields.get_safe_value()).replace(',', ' ').split()
     elif isinstance(_fields, iterable.Sequence):
         fields = [
-            v.get_safe_value()
+            force_unicode(v.get_safe_value())
             for lazy_context in _fields.py__iter__()
             for v in lazy_context.infer() if is_string(v)
         ]