diff --git a/benchmarks/benchmark_kelvin.cc b/benchmarks/benchmark_kelvin.cc
index 7756b34..eac9fe8 100644
--- a/benchmarks/benchmark_kelvin.cc
+++ b/benchmarks/benchmark_kelvin.cc
@@ -52,7 +52,7 @@
 #endif
 
 namespace {
-constexpr int kTensorArenaSize = 1536 * 1024;
+constexpr int kTensorArenaSize = 2 * 1024 * 1024;
 uint8_t g_tensor_arena[kTensorArenaSize] __attribute__((aligned(64)));
 
 __attribute__((section(".model_output_header"))) BenchmarkOutputHeader output_header = {
@@ -63,7 +63,7 @@
 };
 
 // This includes all ops currently used in the Kelvin model suite. More can be added.
-constexpr int kAllOpsNum = 25;
+constexpr int kAllOpsNum = 28;
 std::unique_ptr<tflite::MicroMutableOpResolver<kAllOpsNum>> GetAllOpsResolver() {
   tflite::MicroMutableOpResolver<kAllOpsNum> resolver;
   resolver.AddAveragePool2D();
@@ -91,6 +91,9 @@
   resolver.AddMean();
   resolver.AddPack();
   resolver.AddShape();
+  resolver.AddResizeNearestNeighbor();
+  resolver.AddTranspose();
+  resolver.AddMul();
   return std::make_unique<tflite::MicroMutableOpResolver<kAllOpsNum>>(resolver);
 }