diff --git a/litellm/proxy/client/cli/commands/agents.py b/litellm/proxy/client/cli/commands/agents.py
index 8e3698985af..0354aefa95c 100644
--- a/litellm/proxy/client/cli/commands/agents.py
+++ b/litellm/proxy/client/cli/commands/agents.py
@@ -453,7 +453,11 @@ def _replace_file(path: Path, text: str) -> None:
     path.parent.mkdir(parents=True, exist_ok=True)
     with tempfile.NamedTemporaryFile("w", encoding="utf-8", dir=path.parent, delete=False) as tmp:
         _ = tmp.write(text)
-    os.replace(tmp.name, path)
+    try:
+        os.replace(tmp.name, path)
+    except OSError:
+        Path(tmp.name).unlink(missing_ok=True)
+        raise
 
 
 def codex_model_sync_args(
diff --git a/tests/test_litellm/proxy/client/cli/test_agents.py b/tests/test_litellm/proxy/client/cli/test_agents.py
index 75e42c3eaa0..3020b770e62 100644
--- a/tests/test_litellm/proxy/client/cli/test_agents.py
+++ b/tests/test_litellm/proxy/client/cli/test_agents.py
@@ -481,6 +481,13 @@ class TestCodexModelSync:
         assert isinstance(result, ModelSyncSkipped)
         assert "could not write" in result.reason
 
+    def test_failed_replace_is_reported_and_leaves_no_temp_file(self, tmp_path):
+        (tmp_path / "litellm-models.json").mkdir()
+        _, result = self._sync(self._listing(self._row("m")), tmp_path)
+        assert isinstance(result, ModelSyncSkipped)
+        assert "could not write" in result.reason
+        assert [p.name for p in tmp_path.iterdir()] == ["litellm-models.json"]
+
     def test_unreachable_proxy_is_reported_not_raised(self, tmp_path):
         def boom(*a, **k):
             raise requests.ConnectionError("refused")