diff --git a/tensorflow/lite/tools/benchmark/README.md b/tensorflow/lite/tools/benchmark/README.md
index a4f632c40a9..c44129cbbd3 100644
--- a/tensorflow/lite/tools/benchmark/README.md
+++ b/tensorflow/lite/tools/benchmark/README.md
@@ -87,6 +87,7 @@ the reported data on hexagon is in cycles, not in ms like on cpu.
 
 #### CoreML delegate
 *   `use_coreml`: `bool` (default=false)
+*   `coreml_version`: `int` (default=0)
 
 #### External delegate
 *   `external_delegate_path`: `string` (default="")
diff --git a/tensorflow/lite/tools/delegates/README.md b/tensorflow/lite/tools/delegates/README.md
index f0e15e9e71a..709fcffb24d 100644
--- a/tensorflow/lite/tools/delegates/README.md
+++ b/tensorflow/lite/tools/delegates/README.md
@@ -93,6 +93,9 @@ TFLite delegate.
 *   `use_coreml`: `bool` (default=false) \
     Whether to use the [Core ML delegate](https://github.com/tensorflow/tensorflow/tree/master/tensorflow/lite/experimental/delegates/coreml).
     This option is only available in iOS.
+*   `coreml_version`: `int` (default=0) \
+    Target Core ML version for model conversion. The default value is 0 and it
+    means using the newest version that's available on the device.
 
 ### External delegate provider
 *   `external_delegate_path`: `string` (default="") \
diff --git a/tensorflow/lite/tools/delegates/coreml_delegate_provider.cc b/tensorflow/lite/tools/delegates/coreml_delegate_provider.cc
index 0d1a8ade368..c29555716a4 100644
--- a/tensorflow/lite/tools/delegates/coreml_delegate_provider.cc
+++ b/tensorflow/lite/tools/delegates/coreml_delegate_provider.cc
@@ -32,6 +32,7 @@ class CoreMlDelegateProvider : public DelegateProvider {
   CoreMlDelegateProvider() {
 #if defined(REAL_IPHONE_DEVICE)
     default_params_.AddParam("use_coreml", ToolParam::Create<bool>(true));
+    default_params_.AddParam("coreml_version", ToolParam::Create<int>(0));
 #endif
   }
   std::vector<Flag> CreateFlags(ToolParams* params) const final;
@@ -49,6 +50,10 @@ std::vector<Flag> CoreMlDelegateProvider::CreateFlags(
 #if defined(REAL_IPHONE_DEVICE)
   std::vector<Flag> flags = {
       CreateFlag<bool>("use_coreml", params, "use Core ML"),
+      CreateFlag<int>("coreml_version", params,
+                      "Target Core ML version for model conversion. "
+                      "The default value is 0 and it means using the newest "
+                      "version that's available on the device."),
   };
   return flags;
 #else
@@ -71,6 +76,7 @@ TfLiteDelegatePtr CoreMlDelegateProvider::CreateTfLiteDelegate(
   if (params.Get<bool>("use_coreml")) {
     TfLiteCoreMlDelegateOptions coreml_opts = {
         .enabled_devices = TfLiteCoreMlDelegateAllDevices};
+    coreml_opts.coreml_version = params.Get<int>("coreml_version");
     coreml_opts.max_delegated_partitions =
         params.Get<int>("max_delegated_partitions");
     coreml_opts.min_nodes_per_partition =