diff --git a/src/generators.cpp b/src/generators.cpp
index 388e63501e..cd7eed5497 100644
--- a/src/generators.cpp
+++ b/src/generators.cpp
@@ -380,8 +380,20 @@ Generator::Generator(const Model& model, const GeneratorParams& params) : model_
     throw std::runtime_error("search max_length is 0");
   if (params.search.max_length > model.config_->model.context_length)
     throw std::runtime_error("max_length (" + std::to_string(params.search.max_length) + ") cannot be greater than model context_length (" + std::to_string(model.config_->model.context_length) + ")");
-  if (params.search.batch_size < 1)
-    throw std::runtime_error("batch_size must be 1 or greater, is " + std::to_string(params.search.batch_size));
+
+  constexpr int kMaxBatchSize = 32;
+  constexpr int kMaxNumBeams = 32;
+  constexpr int kMaxNumBeamsCuda = 32;
+
+  if (params.search.batch_size < 1 || params.search.batch_size > kMaxBatchSize)
+    throw std::runtime_error("batch_size (" + std::to_string(params.search.batch_size) + ") must be in [1, " + std::to_string(kMaxBatchSize) + "]");
+
+  const int max_num_beams = (params.search.num_beams > 1 &&
+                             (params.p_device->GetType() == DeviceType::CUDA || params.p_device->GetType() == DeviceType::NvTensorRtRtx))
+                                ? kMaxNumBeamsCuda
+                                : kMaxNumBeams;
+  if (params.search.num_beams < 1 || params.search.num_beams > max_num_beams)
+    throw std::runtime_error("num_beams (" + std::to_string(params.search.num_beams) + ") must be in [1, " + std::to_string(max_num_beams) + "]");
   if (params.config.model.vocab_size < 1)
     throw std::runtime_error("vocab_size must be 1 or greater, is " + std::to_string(params.config.model.vocab_size));
 
diff --git a/test/model_tests.cpp b/test/model_tests.cpp
index a96cd7a7d0..42ebbd53eb 100644
--- a/test/model_tests.cpp
+++ b/test/model_tests.cpp
@@ -482,6 +482,46 @@ Print all primes between 1 and n
 }
 #endif
 
+// Validation tests for search parameter bounds
+#if !USE_DML
+TEST(ModelTests, NumBeamsUpperBoundThrows) {
+  auto model = OgaModel::Create(MODEL_PATH "hf-internal-testing/tiny-random-gpt2-fp32");
+  auto params = OgaGeneratorParams::Create(*model);
+  params->SetSearchOption("max_length", 20);
+  params->SetSearchOption("batch_size", 1);
+  params->SetSearchOption("num_beams", 512);  // exceeds upper bound of 256
+
+  EXPECT_THROW(OgaGenerator::Create(*model, *params), std::runtime_error);
+}
+
+TEST(ModelTests, BatchSizeUpperBoundThrows) {
+  auto model = OgaModel::Create(MODEL_PATH "hf-internal-testing/tiny-random-gpt2-fp32");
+  auto params = OgaGeneratorParams::Create(*model);
+  params->SetSearchOption("max_length", 20);
+  params->SetSearchOption("batch_size", 512);  // exceeds upper bound of 256
+
+  EXPECT_THROW(OgaGenerator::Create(*model, *params), std::runtime_error);
+}
+
+TEST(ModelTests, NumBeamsZeroThrows) {
+  auto model = OgaModel::Create(MODEL_PATH "hf-internal-testing/tiny-random-gpt2-fp32");
+  auto params = OgaGeneratorParams::Create(*model);
+  params->SetSearchOption("max_length", 20);
+  params->SetSearchOption("batch_size", 1);
+  params->SetSearchOption("num_beams", 0);  // below lower bound of 1
+
+  EXPECT_THROW(OgaGenerator::Create(*model, *params), std::runtime_error);
+}
+
+TEST(ModelTests, BatchSizeZeroThrows) {
+  auto model = OgaModel::Create(MODEL_PATH "hf-internal-testing/tiny-random-gpt2-fp32");
+  auto params = OgaGeneratorParams::Create(*model);
+  params->SetSearchOption("max_length", 20);
+  params->SetSearchOption("batch_size", 0);  // below lower bound of 1
+
+  EXPECT_THROW(OgaGenerator::Create(*model, *params), std::runtime_error);
+}
+#endif
 // --- Validation tests (no model files required) ---
 
 TEST(ValidationTests, WindowIndexAcceptsValidParams) {
@@ -520,4 +560,4 @@ TEST(ValidationTests, WindowIndexRejectsExcessiveDimensions) {
 TEST(ValidationTests, WindowIndexRejectsTotalSizeOverflow) {
   // Each dim individually <= kMaxElements, but grid_t * padded_h * padded_w > kMaxElements
   EXPECT_THROW(Generators::ValidateWindowIndexParams(1000, 2000000, 2000000, 2, 14, 112), std::runtime_error);
-}
\ No newline at end of file
+}