diff --git a/jedi/_compatibility.py b/jedi/_compatibility.py
index 5004a7a2..4710e253 100644
--- a/jedi/_compatibility.py
+++ b/jedi/_compatibility.py
@@ -233,6 +233,11 @@ try:
 except ImportError:
     from itertools import izip_longest as zip_longest  # Python 2
 
+try:
+    FileNotFoundError = FileNotFoundError
+except NameError:
+    FileNotFoundError = IOError
+
 
 def no_unicode_pprint(dct):
     """
diff --git a/jedi/parser/__init__.py b/jedi/parser/__init__.py
index 492ca222..c411e878 100644
--- a/jedi/parser/__init__.py
+++ b/jedi/parser/__init__.py
@@ -18,6 +18,7 @@ complexity of the ``Parser`` (there's another parser sitting inside
 import os
 import re
 
+from jedi._compatibility import FileNotFoundError
 from jedi.parser import tree as pt
 from jedi.parser import tokenize
 from jedi.parser.token import (DEDENT, INDENT, ENDMARKER, NEWLINE, NUMBER,
@@ -40,7 +41,7 @@ class ParseError(Exception):
     """
 
 
-def load_grammar(version='3.4'):
+def load_grammar(version='3.6'):
     # For now we only support two different Python syntax versions: The latest
     # Python 3 and Python 2. This may change.
     if version in ('3.2', '3.3'):
@@ -55,7 +56,11 @@ def load_grammar(version='3.4'):
     try:
         return _loaded_grammars[path]
     except KeyError:
-        return _loaded_grammars.setdefault(path, generate_grammar(path))
+        try:
+            return _loaded_grammars.setdefault(path, generate_grammar(path))
+        except FileNotFoundError:
+            # Just load the default if the file does not exist.
+            return load_grammar()
 
 
 class ParserSyntaxError(object):
diff --git a/test/test_parser/test_parser.py b/test/test_parser/test_parser.py
index 14ec763a..09125498 100644
--- a/test/test_parser/test_parser.py
+++ b/test/test_parser/test_parser.py
@@ -234,3 +234,12 @@ def test_python3_octal():
         assert module.children[0].children[0].type == 'number'
     else:
         assert module.children[0].type == 'error_node'
+
+
+def test_load_newer_grammar():
+    # This version shouldn't be out for a while, but if we somehow get this it
+    # should just take the latest Python grammar.
+    load_grammar('15.8')
+    # The same is true for very old grammars (even though this is probably not
+    # going to be an issue.
+    load_grammar('1.5')