diff --git a/paddle/fluid/eager/legacy/prepared_operator.cc b/paddle/fluid/eager/legacy/prepared_operator.cc
index fcdf4162685c5b6668552bfa312016704fb137d3..a9a85d50d6f4ed0c91dd219ffbd9bd0497f07108 100644
--- a/paddle/fluid/eager/legacy/prepared_operator.cc
+++ b/paddle/fluid/eager/legacy/prepared_operator.cc
@@ -24,7 +24,6 @@
 #include "paddle/fluid/platform/device/xpu/xpu_op_list.h"
 #endif
 DECLARE_bool(check_nan_inf);
-DECLARE_bool(run_pten_kernel);
 
 namespace egr {
 namespace legacy {
diff --git a/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cpu.cc b/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cpu.cc
index c100e3b70f3842d3781a461a754c0439919d8a3c..93cd0d1338f684a2d18803f0038c9fb8f53d7dc9 100644
--- a/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cpu.cc
+++ b/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cpu.cc
@@ -36,11 +36,6 @@
 using namespace egr;            // NOLINT
 using namespace egr_utils_api;  // NOLINT
 
-// Disable pten path
-DECLARE_bool(run_pten_kernel);
-
-TEST(Benchmark, Init) { FLAGS_run_pten_kernel = false; }
-
 TEST(Benchmark, EagerScaleCPU) {
   // Prepare Device Contexts
   eager_test::InitEnv(paddle::platform::CPUPlace());
diff --git a/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cuda.cc b/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cuda.cc
index c8f4b1b32e453d85232b5a7a87e8a979aa0637d4..2df44bfcab57fcda293841af8e3a89fa6290499d 100644
--- a/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cuda.cc
+++ b/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cuda.cc
@@ -35,10 +35,6 @@
 using namespace egr;            // NOLINT
 using namespace egr_utils_api;  // NOLINT
 
-DECLARE_bool(run_pten_kernel);
-
-TEST(Benchmark, Init) { FLAGS_run_pten_kernel = false; }
-
 #if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
 
 TEST(Benchmark, EagerScaleCUDA) {
diff --git a/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cpu.cc b/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cpu.cc
index 68e7512eedbde369929ef417e205fdc45bd1ce38..b2a96468ece22c979cbe022531bff9e7739e5153 100644
--- a/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cpu.cc
+++ b/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cpu.cc
@@ -34,11 +34,6 @@
 #include "gperftools/profiler.h"
 #endif
 
-// Disable pten path
-DECLARE_bool(run_pten_kernel);
-
-TEST(Benchmark, Init) { FLAGS_run_pten_kernel = false; }
-
 namespace paddle {
 namespace imperative {
 
diff --git a/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cuda.cc b/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cuda.cc
index 50423b5a64fcf965f50c3d7fc8ec935585e189d4..7f8b845b0703b70523c8732737d182357a64cf83 100644
--- a/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cuda.cc
+++ b/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cuda.cc
@@ -34,11 +34,6 @@
 #include "gperftools/profiler.h"
 #endif
 
-// Disable pten path
-DECLARE_bool(run_pten_kernel);
-
-TEST(Benchmark, Init) { FLAGS_run_pten_kernel = false; }
-
 #if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
 
 namespace paddle {
diff --git a/paddle/fluid/framework/custom_kernel.cc b/paddle/fluid/framework/custom_kernel.cc
index 1ba7c4481edb1b29ec7be45db1efadfd33bf7e07..a04838c9d4f8744cdeeaadb0fdd57c9506f269c5 100644
--- a/paddle/fluid/framework/custom_kernel.cc
+++ b/paddle/fluid/framework/custom_kernel.cc
@@ -29,8 +29,6 @@ limitations under the License. */
 #include "paddle/pten/core/kernel_context.h"
 #include "paddle/pten/core/kernel_registry.h"
 
-DECLARE_bool(run_pten_kernel);
-
 namespace paddle {
 
 namespace framework {
@@ -279,10 +277,6 @@ static void RunKernelFunc(pten::KernelContext* ctx,
 
 void RegisterKernelWithMetaInfo(
     const std::vector<OpKernelInfo>& op_kernel_infos) {
-  PADDLE_ENFORCE_EQ(FLAGS_run_pten_kernel, true,
-                    platform::errors::Unimplemented(
-                        "Custom Kernel depends on pten kernel enabled,"));
-
   for (size_t i = 0; i < op_kernel_infos.size(); ++i) {
     auto& kernel_info = op_kernel_infos[i];
     auto op_type = OpKernelInfoHelper::GetOpName(kernel_info);
diff --git a/paddle/fluid/framework/new_executor/interpretercore_util.cc b/paddle/fluid/framework/new_executor/interpretercore_util.cc
index fb0951e87aa168e0a0277953f3c1b184c6fcaa58..50725e5eea5f1c1b2248019548990f9873ab862e 100644
--- a/paddle/fluid/framework/new_executor/interpretercore_util.cc
+++ b/paddle/fluid/framework/new_executor/interpretercore_util.cc
@@ -24,7 +24,6 @@
 PADDLE_DEFINE_EXPORTED_bool(
     new_executor_sequential_run, false,
     "Enable sequential execution for standalone executor, used for debug");
-DECLARE_bool(run_pten_kernel);
 
 namespace paddle {
 namespace framework {
diff --git a/paddle/fluid/framework/operator.cc b/paddle/fluid/framework/operator.cc
index 2142d10bc43417913ce8ce7022720aa81193fd94..50c315bf03848966d196dd86d2122b346c9e88d4 100644
--- a/paddle/fluid/framework/operator.cc
+++ b/paddle/fluid/framework/operator.cc
@@ -56,7 +56,6 @@ DECLARE_bool(check_nan_inf);
 DECLARE_bool(enable_unused_var_check);
 PADDLE_DEFINE_EXPORTED_int32(inner_op_parallelism, 0,
                              "number of threads for inner op");
-DECLARE_bool(run_pten_kernel);
 DECLARE_bool(run_kp_kernel);
 
 namespace paddle {
diff --git a/paddle/fluid/imperative/prepared_operator.cc b/paddle/fluid/imperative/prepared_operator.cc
index 8eca35c5cec02f9241d10a17935eb0a96ecc2475..53f5e0ca8fd378b3f456ab934de68c4c9e2e4efa 100644
--- a/paddle/fluid/imperative/prepared_operator.cc
+++ b/paddle/fluid/imperative/prepared_operator.cc
@@ -29,7 +29,6 @@
 #include "paddle/fluid/platform/profiler.h"
 
 DECLARE_bool(check_nan_inf);
-DECLARE_bool(run_pten_kernel);
 DECLARE_bool(benchmark);
 DECLARE_bool(run_kp_kernel);
 
diff --git a/paddle/fluid/platform/flags.cc b/paddle/fluid/platform/flags.cc
index d195b5c04ace71178abd42a6279520d3b26a6029..b969ba971b6b1ec2ca1ad6e8c0c28fdf07bb6431 100644
--- a/paddle/fluid/platform/flags.cc
+++ b/paddle/fluid/platform/flags.cc
@@ -687,18 +687,6 @@ PADDLE_DEFINE_EXPORTED_bool(
     apply_pass_to_program, false,
     "It controls whether to apply IR pass to program when using Fleet APIs");
 
-/**
- * Pt kernel related FLAG
- * Name: FLAGS_run_pten_kernel
- * Since Version: 2.3.0
- * Value Range: bool, default=false
- * Example: FLAGS_run_pten_kernel=true would use the pt kernel to compute in the
- * Op.
- * Note:
- */
-PADDLE_DEFINE_EXPORTED_bool(run_pten_kernel, true,
-                            "It controls whether to use pten kernel");
-
 /**
  * KP kernel related FLAG
  * Name: FLAGS_run_kp_kernel