diff --git a/press/press/doctype/bench/bench.py b/press/press/doctype/bench/bench.py
index 0ca78ef210..8c91b9348a 100644
--- a/press/press/doctype/bench/bench.py
+++ b/press/press/doctype/bench/bench.py
@@ -799,6 +799,11 @@ def allocate_workers(
 					"min_background_workers",
 				),
 			)[0]
+			is_public = frappe.get_cached_value("Server", self.server, "public")
+			if not is_public:
+				bench_config = json.loads(self.bench_config or "{}")
+				min_gn = bench_config.get("min_gunicorn_workers") or min_gn
+				min_bg = bench_config.get("min_background_workers") or min_bg
 			self.gunicorn_workers = min(
 				max_gn or MAX_GUNICORN_WORKERS,
 				max(
diff --git a/press/press/doctype/bench/test_bench.py b/press/press/doctype/bench/test_bench.py
index 2f1923af82..53b6f59508 100644
--- a/press/press/doctype/bench/test_bench.py
+++ b/press/press/doctype/bench/test_bench.py
@@ -287,6 +287,29 @@ def test_auto_scale_uses_release_groups_max_and_min_workers_when_set(self):
 		self.assertEqual(bench.gunicorn_workers, 10)
 		self.assertEqual(bench.background_workers, 5)
 
+	def test_auto_scale_uses_release_groups_max_and_min_workers_when_set_at_bench_config(self):
+		bench = self._create_bench_with_n_sites_with_cpu_time(3, 5)
+		self.assertEqual(bench.gunicorn_workers, 2)
+		self.assertEqual(bench.background_workers, 1)
+		group = frappe.get_doc("Release Group", bench.group)
+		group.db_set("max_gunicorn_workers", 10)
+		group.db_set("max_background_workers", 5)
+		bench.bench_config = json.dumps({"max_gunicorn_workers": 20, "max_background_workers": 10})  # bench config should not override max workers
+		scale_workers()
+		bench.reload()
+		self.assertEqual(bench.gunicorn_workers, 10)
+		self.assertEqual(bench.background_workers, 5)
+		frappe.db.set_value("Server", bench.server, "ram", 1600)
+		scale_workers()
+		bench.reload()
+		self.assertEqual(bench.gunicorn_workers, 5)  # autoscaled for for such low ram
+		self.assertEqual(bench.background_workers, 2)
+		bench.bench_config = json.dumps({"min_gunicorn_workers": 8, "min_background_workers": 4})
+		scale_workers()
+		bench.reload()
+		self.assertEqual(bench.gunicorn_workers, 8)
+		self.assertEqual(bench.background_workers, 4)
+
 	def test_set_bench_memory_limits_on_server_adds_memory_limit_on_bench_on_auto_scale(
 		self,
 	):