diff --git a/sqlmesh/core/dialect.py b/sqlmesh/core/dialect.py
index 1a42480c13..f5464e12bc 100644
--- a/sqlmesh/core/dialect.py
+++ b/sqlmesh/core/dialect.py
@@ -13,6 +13,7 @@
 from sqlglot import Dialect, Generator, ParseError, Parser, Tokenizer, TokenType, exp
 from sqlglot.dialects.dialect import DialectType
 from sqlglot.dialects import DuckDB, Snowflake
+import sqlglot.dialects.athena as athena
 from sqlglot.helper import seq_get
 from sqlglot.optimizer.normalize_identifiers import normalize_identifiers
 from sqlglot.optimizer.qualify_columns import quote_identifiers
@@ -1014,6 +1015,14 @@ def extend_sqlglot() -> None:
     generators = {Generator}
 
     for dialect in Dialect.classes.values():
+        # Athena picks a different Tokenizer / Parser / Generator depending on the query
+        # so this ensures that the extra ones it defines are also extended
+        if dialect == athena.Athena:
+            tokenizers.add(athena._TrinoTokenizer)
+            parsers.add(athena._TrinoParser)
+            generators.add(athena._TrinoGenerator)
+            generators.add(athena._HiveGenerator)
+
         if hasattr(dialect, "Tokenizer"):
             tokenizers.add(dialect.Tokenizer)
         if hasattr(dialect, "Parser"):
diff --git a/tests/core/test_dialect.py b/tests/core/test_dialect.py
index ebf90bebf7..11ffec3720 100644
--- a/tests/core/test_dialect.py
+++ b/tests/core/test_dialect.py
@@ -12,7 +12,9 @@
     select_from_values_for_batch_range,
     text_diff,
 )
+import sqlmesh.core.dialect as d
 from sqlmesh.core.model import SqlModel, load_sql_based_model
+from sqlmesh.core.config.connection import DIALECT_TO_TYPE
 
 
 def test_format_model_expressions():
@@ -700,3 +702,18 @@ def test_model_name_cannot_be_string():
 
 def test_parse_snowflake_create_schema_ddl():
     assert parse_one("CREATE SCHEMA d.s", dialect="snowflake").sql() == "CREATE SCHEMA d.s"
+
+
+@pytest.mark.parametrize("dialect", sorted(set(DIALECT_TO_TYPE.values())))
+def test_sqlglot_extended_correctly(dialect: str) -> None:
+    # MODEL is a SQLMesh extension and not part of SQLGlot
+    # If we can roundtrip an expression containing MODEL across every dialect, then the SQLMesh extensions have been registered correctly
+    ast = d.parse_one("MODEL (name foo)", dialect=dialect)
+    assert isinstance(ast, d.Model)
+    name_prop = ast.find(exp.Property)
+    assert isinstance(name_prop, exp.Property)
+    assert name_prop.this == "name"
+    value = name_prop.args["value"]
+    assert isinstance(value, exp.Table)
+    assert value.sql() == "foo"
+    assert ast.sql(dialect=dialect) == "MODEL (\nname foo\n)"