diff --git a/scripts/generate_wrappers.py b/scripts/generate_wrappers.py
index 88d1ba186..c4c0ee1a8 100644
--- a/scripts/generate_wrappers.py
+++ b/scripts/generate_wrappers.py
@@ -1534,6 +1534,13 @@ def _append_unique(declaration, definition):
         declarations.append(declaration)
         definitions.append(definition)
 
+    _append_unique(
+        f"extern template std::size_t "
+        f"Operator<{op_type}>::DefaultImplementationIndex(Device::Type);",
+        f"template std::size_t "
+        f"Operator<{op_type}>::DefaultImplementationIndex(Device::Type);",
+    )
+
     for call in operator.calls:
         template_arguments = _generate_template_arguments(call)
         params = _generate_parameters(call)
diff --git a/src/operator.h b/src/operator.h
index c3c98e00f..1c60596a4 100644
--- a/src/operator.h
+++ b/src/operator.h
@@ -3,6 +3,7 @@
 
 #include <atomic>
 #include <cassert>
+#include <cstdlib>
 #include <memory>
 #include <optional>
 #include <tuple>
@@ -206,7 +207,8 @@ class Operator : public OperatorBase {
 
   template <typename... Args>
   static std::unique_ptr<Operator> Make(const Tensor tensor, Args&&... args) {
-    return Make({}, tensor, std::forward<Args>(args)...);
+    return Make(DefaultConfig(tensor.device().type()), tensor,
+                std::as_const(args)...);
   }
 
   template <typename... Args>
@@ -222,7 +224,10 @@ class Operator : public OperatorBase {
   template <typename... Args>
   static std::unique_ptr<Operator> Make(const std::vector<Tensor> tensors,
                                         Args&&... args) {
-    return Make({}, tensors, std::forward<Args>(args)...);
+    assert(!tensors.empty() && "operator tensor list input cannot be empty");
+
+    return Make(DefaultConfig(tensors.front().device().type()), tensors,
+                std::as_const(args)...);
   }
 
   template <typename... Args>
@@ -279,7 +284,7 @@ class Operator : public OperatorBase {
 
   template <typename... Args>
   static void Call(const Tensor tensor, const Args&... args) {
-    return Call({}, {}, tensor, args...);
+    return Call({}, DefaultConfig(tensor.device().type()), tensor, args...);
   }
 
   template <
@@ -329,6 +334,39 @@ class Operator : public OperatorBase {
   static constexpr std::size_t implementation_index_{implementation_index};
 
  private:
+  template <auto first, auto... rest>
+  static constexpr std::size_t FirstActiveImplementationIndex(
+      List<first, rest...>) {
+    return static_cast<std::size_t>(first);
+  }
+
+  static std::size_t FirstActiveImplementationIndex(List<>) {
+    assert(false && "operator has no active implementation for this device");
+    std::abort();
+  }
+
+  static std::size_t DefaultImplementationIndex(Device::Type dev_type) {
+    std::size_t default_index{0};
+
+    DispatchFunc<ActiveDevices<Key>>(
+        dev_type,
+        [&](auto device_tag) {
+          constexpr Device::Type kDev = decltype(device_tag)::value;
+          default_index = FirstActiveImplementationIndex(
+              typename ActiveImplementations<Key, kDev>::type{});
+        },
+        "Operator::DefaultImplementationIndex");
+
+    return default_index;
+  }
+
+  static Config DefaultConfig(Device::Type dev_type) {
+    Config config;
+    config.set_implementation_index(DefaultImplementationIndex(dev_type));
+
+    return config;
+  }
+
   template <typename TensorLike, typename... Args>
   static auto CallReturning(const TensorLike& tensor, const Args&... args) {
     auto out = Key::MakeReturnValue(tensor, args...);
diff --git a/tests/test_cpp_api.py b/tests/test_cpp_api.py
index 1cc60fe34..52a97542f 100644
--- a/tests/test_cpp_api.py
+++ b/tests/test_cpp_api.py
@@ -58,6 +58,32 @@ def test_cpp_returning_call_smoke(tmp_path):
     _run([str(binary)])
 
 
+def test_cpp_configless_calls_use_first_active_implementation(tmp_path):
+    install_prefix = _install_prefix()
+    include_dir = install_prefix / "include"
+    library_dir = _library_dir(install_prefix)
+    source = tmp_path / "configless_active_implementation.cc"
+    binary = tmp_path / "configless_active_implementation"
+    source.write_text(_CONFIGLESS_ACTIVE_IMPLEMENTATION_SOURCE)
+
+    _run(
+        [
+            _compiler("CXX", "c++"),
+            "-std=c++17",
+            "-Werror",
+            f"-I{include_dir}",
+            str(source),
+            f"-L{library_dir}",
+            "-linfiniops",
+            "-linfinirt",
+            f"-Wl,-rpath,{library_dir}",
+            "-o",
+            str(binary),
+        ]
+    )
+    _run([str(binary)])
+
+
 @pytest.mark.parametrize(
     "header",
     (
@@ -332,3 +358,119 @@ class OwningTensor {
     }
     """
 ).lstrip()
+
+
+_CONFIGLESS_ACTIVE_IMPLEMENTATION_SOURCE = textwrap.dedent(
+    r"""
+    #include <infini/ops.h>
+
+    #include <cmath>
+    #include <cstdint>
+    #include <vector>
+
+    namespace infini::ops {
+
+    class ConfiglessSelection : public Operator<ConfiglessSelection> {
+     public:
+      ConfiglessSelection(const Tensor input, Tensor out) {}
+
+      ConfiglessSelection(const std::vector<Tensor> inputs, Tensor out) {}
+
+      virtual void operator()(const Tensor input, Tensor out) const = 0;
+
+      virtual void operator()(const std::vector<Tensor> inputs,
+                              Tensor out) const = 0;
+
+      template <typename TensorLike>
+      static auto MakeReturnValue(const TensorLike& input) {
+        return TensorLike::Empty(input.shape(), input.dtype(), input.device());
+      }
+    };
+
+    template <>
+    class Operator<ConfiglessSelection, Device::Type::kCpu, 16>
+        : public ConfiglessSelection {
+     public:
+      using ConfiglessSelection::ConfiglessSelection;
+
+      void operator()(const Tensor input, Tensor out) const override {
+        const auto* input_data = static_cast<const float*>(input.data());
+        auto* out_data = static_cast<float*>(out.data());
+        out_data[0] = input_data[0] + 16.0f;
+      }
+
+      void operator()(const std::vector<Tensor> inputs,
+                      Tensor out) const override {
+        const auto* first_data = static_cast<const float*>(inputs[0].data());
+        const auto* second_data = static_cast<const float*>(inputs[1].data());
+        auto* out_data = static_cast<float*>(out.data());
+        out_data[0] = first_data[0] + second_data[0] + 16.0f;
+      }
+    };
+
+    }  // namespace infini::ops
+
+    int main() {
+      float input_data = 1.0f;
+      float second_data = 2.0f;
+      float out_data = 0.0f;
+      const infini::ops::Tensor::Shape shape{1};
+      const infini::ops::Device device{infini::ops::Device::Type::kCpu};
+      const infini::ops::DataType dtype{infini::ops::DataType::kFloat32};
+      infini::ops::Tensor input(&input_data, shape, dtype, device);
+      infini::ops::Tensor second(&second_data, shape, dtype, device);
+      infini::ops::Tensor out(&out_data, shape, dtype, device);
+
+      auto tensor_op = infini::ops::ConfiglessSelection::Make(input, out);
+      (*tensor_op)(input, out);
+      if (std::fabs(out_data - 17.0f) > 1e-6f) {
+        return 1;
+      }
+
+      out_data = 0.0f;
+      std::vector<infini::ops::Tensor> inputs{input, second};
+      auto vector_op = infini::ops::ConfiglessSelection::Make(inputs, out);
+      (*vector_op)(inputs, out);
+      if (std::fabs(out_data - 19.0f) > 1e-6f) {
+        return 2;
+      }
+
+      float cat_out_data[2] = {};
+      const infini::ops::Tensor::Shape cat_shape{2};
+      infini::ops::Tensor cat_out(cat_out_data, cat_shape, dtype, device);
+      auto cat_op =
+          infini::ops::Cat::Make(inputs, std::int64_t{0}, cat_out);
+      (*cat_op)(inputs, std::int64_t{0}, cat_out);
+      if (std::fabs(cat_out_data[0] - 1.0f) > 1e-6f ||
+          std::fabs(cat_out_data[1] - 2.0f) > 1e-6f) {
+        return 3;
+      }
+
+      float abs_input_data = -4.0f;
+      float abs_out_data = 0.0f;
+      infini::ops::Tensor abs_input(&abs_input_data, shape, dtype, device);
+      infini::ops::Tensor abs_out(&abs_out_data, shape, dtype, device);
+      auto abs_op = infini::ops::Abs::Make(abs_input, abs_out);
+      (*abs_op)(abs_input, abs_out);
+      if (std::fabs(abs_out_data - 4.0f) > 1e-6f) {
+        return 4;
+      }
+
+      auto abs_op_from_rvalue = infini::ops::Abs::Make(
+          abs_input,
+          infini::ops::Tensor(&abs_out_data, shape, dtype, device));
+      (*abs_op_from_rvalue)(abs_input, abs_out);
+      if (std::fabs(abs_out_data - 4.0f) > 1e-6f) {
+        return 5;
+      }
+
+      abs_out_data = 0.0f;
+      infini::ops::Abs::Call(abs_input, abs_out);
+      if (std::fabs(abs_out_data - 4.0f) > 1e-6f) {
+        return 6;
+      }
+
+      return 0;
+    }
+    """
+).lstrip()
diff --git a/tests/test_generate_wrappers.py b/tests/test_generate_wrappers.py
index f347d2964..998109de6 100644
--- a/tests/test_generate_wrappers.py
+++ b/tests/test_generate_wrappers.py
@@ -78,6 +78,33 @@ class Clamp {
     ) in text
 
 
+def test_operator_call_instantiations_externalize_default_implementation_lookup():
+    module = _load_generator_module()
+    operator = module._Operator(
+        "abs",
+        constructors=[],
+        calls=[
+            module._ParsedFunction(
+                [
+                    module._ParsedArgument("const Tensor", "input"),
+                    module._ParsedArgument("Tensor", "out"),
+                ]
+            )
+        ],
+    )
+
+    declarations, definitions = module._generate_operator_call_instantiation_entries(
+        operator
+    )
+
+    signature = (
+        "std::size_t "
+        "Operator<::infini::ops::Abs>::DefaultImplementationIndex(Device::Type);"
+    )
+    assert f"extern template {signature}" in declarations
+    assert f"template {signature}" in definitions
+
+
 def test_operator_call_instantiations_keep_scalar_and_optional_tensor_overloads_distinct(
     monkeypatch, tmp_path
 ):