diff --git a/RELEASE.md b/RELEASE.md
index ce7830667da..c31c105919c 100644
--- a/RELEASE.md
+++ b/RELEASE.md
@@ -8,6 +8,14 @@
 
 * <DOCUMENT BREAKING CHANGES HERE>
 * <THIS SECTION SHOULD CONTAIN API, ABI AND BEHAVIORAL BREAKING CHANGES>
+* `LiteRT`, a.k.a. `tf.lite`:
+  * C++ API:
+    * The public constants `tflite::Interpreter:kTensorsReservedCapacity`
+      and `tflite::Interpreter:kTensorsCapacityHeadroom` are now const
+      references, rather than `constexpr` compile-time constants.
+      (This is to enable better API compatibility for TFLite in Play services
+      while preserving the implementation flexibility to change the values of
+      these constants in the future.)
 
 ### Known Caveats
 
diff --git a/tensorflow/lite/core/interpreter.cc b/tensorflow/lite/core/interpreter.cc
index 00f1a93ec65..c722beeafc4 100644
--- a/tensorflow/lite/core/interpreter.cc
+++ b/tensorflow/lite/core/interpreter.cc
@@ -66,6 +66,9 @@ static_assert(sizeof(TfLiteFloat16) == sizeof(uint16_t),
 
 namespace tflite {
 
+const int& Interpreter::kTensorsReservedCapacity = 128;
+const int& Interpreter::kTensorsCapacityHeadroom = 16;
+
 namespace {
 
 // Gets the current TfLiteQuantization from the legacy TfLiteQuantizationParams.
diff --git a/tensorflow/lite/core/interpreter.h b/tensorflow/lite/core/interpreter.h
index feb4995b0e9..9c233ab275a 100644
--- a/tensorflow/lite/core/interpreter.h
+++ b/tensorflow/lite/core/interpreter.h
@@ -693,13 +693,14 @@ class Interpreter {
   /// \brief Gets the profiler used for op tracing.
   Profiler* GetProfiler();
 
-  // The default capacity of `tensors_` vector.
-  static constexpr int kTensorsReservedCapacity = 128;
-  /// The capacity headroom of `tensors_` vector before calling ops'
-  /// `prepare` and `invoke` function. In these functions, it's guaranteed
-  /// allocating up to `kTensorsCapacityHeadroom` more tensors won't invalidate
+  /// The default capacity of the tensors vector.
+  static const int& kTensorsReservedCapacity;
+
+  /// The capacity headroom of the tensors vector before calling ops'
+  /// `prepare` and `invoke` function. In those functions, it's guaranteed
+  /// allocating up to this many more tensors won't invalidate
   /// pointers to existing tensors.
-  static constexpr int kTensorsCapacityHeadroom = 16;
+  static const int& kTensorsCapacityHeadroom;
 
   /// \warning This is an experimental API and subject to change. \n
   /// \brief Set if buffer handle output is allowed.