diff --git a/src/databricks/sql/auth/common.py b/src/databricks/sql/auth/common.py
index cda30a528..a0976e78c 100644
--- a/src/databricks/sql/auth/common.py
+++ b/src/databricks/sql/auth/common.py
@@ -73,11 +73,16 @@ def __init__(
         # HTTP client configuration
         self.ssl_options = ssl_options
         self.socket_timeout = socket_timeout
-        self.retry_stop_after_attempts_count = retry_stop_after_attempts_count or 5
+        # Defaults align with the connector-wide retry defaults used by the
+        # Thrift and SEA backends (30 attempts / 900.0s), which are anchored to
+        # the ODBC/JDBC drivers. UnifiedHttpClient (used by CloudFetch downloads)
+        # builds its retry policy from this context, so diverging here cut
+        # cloudfetch retries off far too early (see issue #709).
+        self.retry_stop_after_attempts_count = retry_stop_after_attempts_count or 30
         self.retry_delay_min = retry_delay_min or 1.0
         self.retry_delay_max = retry_delay_max or 10.0
         self.retry_stop_after_attempts_duration = (
-            retry_stop_after_attempts_duration or 300.0
+            retry_stop_after_attempts_duration or 900.0
         )
         self.retry_delay_default = retry_delay_default or 5.0
         self.retry_dangerous_codes = retry_dangerous_codes or []
diff --git a/tests/unit/test_client_context_retry_defaults.py b/tests/unit/test_client_context_retry_defaults.py
new file mode 100644
index 000000000..ccc5a7da8
--- /dev/null
+++ b/tests/unit/test_client_context_retry_defaults.py
@@ -0,0 +1,38 @@
+"""
+Unit tests for ClientContext / build_client_context retry-policy defaults.
+
+Regression tests for issue #709: CloudFetch downloads go through
+UnifiedHttpClient, which builds its DatabricksRetryPolicy from ClientContext.
+When the user does not override the retry settings, those defaults must match
+the connector-wide defaults used by the Thrift and SEA backends (30 attempts /
+900.0s), which are anchored to the ODBC/JDBC drivers (see
+thrift_backend.py: "900s attempts-duration lines up w ODBC/JDBC drivers").
+Previously ClientContext fell back to 5 attempts / 300.0s, cutting cloudfetch
+retries off far earlier than the rest of the connector.
+"""
+
+from databricks.sql.auth.common import ClientContext
+from databricks.sql.utils import build_client_context
+
+
+class TestClientContextRetryDefaults:
+    def test_default_max_retry_duration_is_900(self):
+        """With no override, retry duration defaults to 900.0s (not 300.0)."""
+        context = build_client_context("test.databricks.com", "test-version")
+        assert context.retry_stop_after_attempts_duration == 900.0
+
+    def test_default_max_retry_count_is_30(self):
+        """With no override, retry attempt count defaults to 30 (not 5)."""
+        context = build_client_context("test.databricks.com", "test-version")
+        assert context.retry_stop_after_attempts_count == 30
+
+    def test_user_override_still_honored(self):
+        """Explicit user overrides are propagated unchanged."""
+        context = build_client_context(
+            "test.databricks.com",
+            "test-version",
+            _retry_stop_after_attempts_duration=120.0,
+            _retry_stop_after_attempts_count=7,
+        )
+        assert context.retry_stop_after_attempts_duration == 120.0
+        assert context.retry_stop_after_attempts_count == 7