diff --git a/samples/cpp/benchmark_app/utils.cpp b/samples/cpp/benchmark_app/utils.cpp
index 51d0443cfdc..fbdee288211 100644
--- a/samples/cpp/benchmark_app/utils.cpp
+++ b/samples/cpp/benchmark_app/utils.cpp
@@ -112,7 +112,8 @@ std::vector<std::string> parse_devices(const std::string& device_string) {
         auto bracket = comma_separated_devices.find("(");  // e.g. in BATCH:GPU(4)
         comma_separated_devices = comma_separated_devices.substr(colon + 1, bracket - colon - 1);
     }
-    if ((comma_separated_devices == "MULTI") || (comma_separated_devices == "HETERO"))
+    if ((comma_separated_devices == "AUTO") || (comma_separated_devices == "MULTI") ||
+        (comma_separated_devices == "HETERO"))
         return std::vector<std::string>();
     auto devices = split(comma_separated_devices, ',');
     return devices;
diff --git a/src/plugins/auto/executable_network.cpp b/src/plugins/auto/executable_network.cpp
index 303c0d43dae..e351d052ce6 100644
--- a/src/plugins/auto/executable_network.cpp
+++ b/src/plugins/auto/executable_network.cpp
@@ -309,6 +309,29 @@ void MultiDeviceExecutableNetwork::TryToLoadNetWork(AutoLoadContext& context,
     auto& deviceConfig = context.deviceInfo.config;
     auto& deviceList = context.metaDevices;
     bool curDevIsCPU = (device.find("CPU") != std::string::npos);
+    bool curDevIsGPU = (device.find("GPU") != std::string::npos);
+    {
+        std::lock_guard<std::mutex> lock(_confMutex);
+        if (curDevIsGPU && _loadContext[CPU].isEnabled) {
+            // user does not set the compiling threads
+            // limit the threads num for compiling
+            int maxNumThreads = 0;
+            try {
+                maxNumThreads = _core->GetConfig(device, GPU_CONFIG_KEY(MAX_NUM_THREADS)).as<int>();
+            } catch (...) {
+                LOG_DEBUG("[AUTOPLUGIN]: cannot get MAX_NUM_THREADS from GPU");
+            }
+            if (maxNumThreads == static_cast<int>(std::thread::hardware_concurrency())) {
+                int threadNum = maxNumThreads / 2;
+                deviceConfig[GPU_CONFIG_KEY(MAX_NUM_THREADS)] = std::to_string(threadNum).c_str();
+                LOG_DEBUG("[AUTO PLUGIN]:gpu streams number for compiling: %s", deviceConfig[GPU_CONFIG_KEY(MAX_NUM_THREADS)].c_str());
+            } else {
+                // user set the compiling threads num
+                // use the user's val anyway
+                LOG_DEBUG("[AUTOPLUGIN]:user defined compiling threads: %d", maxNumThreads);
+            }
+        }
+    }
     try {
         if (!modelPath.empty()) {
             context.executableNetwork = _core->LoadNetwork(modelPath, device, deviceConfig);
diff --git a/src/tests/unit/auto/auto_release_helper_test.cpp b/src/tests/unit/auto/auto_release_helper_test.cpp
index fee972400af..f3c97ae5ba3 100644
--- a/src/tests/unit/auto/auto_release_helper_test.cpp
+++ b/src/tests/unit/auto/auto_release_helper_test.cpp
@@ -118,6 +118,8 @@ public:
        IE_SET_METRIC(SUPPORTED_CONFIG_KEYS, supportConfigs, {});
        ON_CALL(*core, GetMetric(_, StrEq(METRIC_KEY(SUPPORTED_CONFIG_KEYS)), _))
            .WillByDefault(Return(supportConfigs));
+       ON_CALL(*core, GetConfig(_, StrEq(GPU_CONFIG_KEY(MAX_NUM_THREADS))))
+           .WillByDefault(Return(12));
     }
 };
 
diff --git a/src/tests/unit/auto/auto_select_device_failed_test.cpp b/src/tests/unit/auto/auto_select_device_failed_test.cpp
index eca8e49b8fd..29992dd74c9 100644
--- a/src/tests/unit/auto/auto_select_device_failed_test.cpp
+++ b/src/tests/unit/auto/auto_select_device_failed_test.cpp
@@ -161,6 +161,8 @@ public:
        ON_CALL(*core, GetMetric(_, StrEq(METRIC_KEY(SUPPORTED_CONFIG_KEYS)), _))
            .WillByDefault(Return(supportConfigs));
        EXPECT_CALL(*core, GetMetric(_, StrEq(METRIC_KEY(SUPPORTED_CONFIG_KEYS)), _)).Times(AnyNumber());
+       ON_CALL(*core, GetConfig(_, StrEq(GPU_CONFIG_KEY(MAX_NUM_THREADS))))
+           .WillByDefault(Return(12));
     }
 };
 
diff --git a/src/tests/unit/auto/exec_network_get_metrics.cpp b/src/tests/unit/auto/exec_network_get_metrics.cpp
index acd6af8ad31..80d0d1deb86 100644
--- a/src/tests/unit/auto/exec_network_get_metrics.cpp
+++ b/src/tests/unit/auto/exec_network_get_metrics.cpp
@@ -202,6 +202,9 @@ TEST_P(ExecNetworkGetMetric, OPTIMAL_NUMBER_OF_INFER_REQUESTS) {
         ON_CALL(*core.get(), GetConfig(_, StrEq(CONFIG_KEY(PERFORMANCE_HINT_NUM_REQUESTS))))
             .WillByDefault(Return(std::to_string(gpuPerfHintNum)));
         EXPECT_CALL(*core.get(), GetConfig(_, StrEq(CONFIG_KEY(PERFORMANCE_HINT_NUM_REQUESTS)))).Times(AnyNumber());
+        ON_CALL(*core, GetConfig(_, StrEq(GPU_CONFIG_KEY(MAX_NUM_THREADS))))
+           .WillByDefault(Return(8));
+        EXPECT_CALL(*core.get(), GetConfig(_, StrEq(GPU_CONFIG_KEY(MAX_NUM_THREADS)))).Times(AnyNumber());
     } else {
         metaDevices.push_back({CommonTestUtils::DEVICE_CPU, {}, cpuCustomerNum, ""});
         metaDevices.push_back({actualDeviceName, {}, actualCustomerNum, ""});