diff --git a/src/frontends/paddle/src/frontend.cpp b/src/frontends/paddle/src/frontend.cpp
index 9582fccf6c447f..8caf37968db985 100644
--- a/src/frontends/paddle/src/frontend.cpp
+++ b/src/frontends/paddle/src/frontend.cpp
@@ -135,22 +135,32 @@ bool normalize_framework_node(const std::shared_ptr<FrameworkNode>& node,
     return true;
 }
 
-std::istream* variant_to_stream_ptr(const ov::Any& variant, std::ifstream& ext_stream) {
+OPENVINO_SUPPRESS_DEPRECATED_START
+std::istream* variant_to_stream_ptr(const ov::Any& variant, std::fstream& fs, std::stringstream& ss) {
     if (variant.is<std::istream*>()) {
         return variant.as<std::istream*>();
+    } else if (variant.is<std::shared_ptr<ngraph::runtime::AlignedBuffer>>()) {
+        auto& aligned_weights_buffer = variant.as<std::shared_ptr<ngraph::runtime::AlignedBuffer>>();
+        ss.write(aligned_weights_buffer->get_ptr<char>(), aligned_weights_buffer->size());
+        FRONT_END_INITIALIZATION_CHECK(ss && ss.good(), "Cannot open ov::tensor.");
+        return &ss;
     } else if (variant.is<std::string>()) {
         const auto& model_path = variant.as<std::string>();
-        ext_stream.open(model_path, std::ios::in | std::ifstream::binary);
+        fs.open(model_path, std::ios::in | std::ifstream::binary);
+        FRONT_END_INITIALIZATION_CHECK(fs && fs.is_open(), "Cannot open model file.");
+        return &fs;
     }
 #if defined(OPENVINO_ENABLE_UNICODE_PATH_SUPPORT) && defined(_WIN32)
     else if (variant.is<std::wstring>()) {
         const auto& model_path = variant.as<std::wstring>();
-        ext_stream.open(model_path.c_str(), std::ios::in | std::ifstream::binary);
+        fs.open(model_path.c_str(), std::ios::in | std::ifstream::binary);
+        FRONT_END_INITIALIZATION_CHECK(fs && fs.is_open(), "Cannot open model file.");
+        return &fs;
     }
 #endif
-    FRONT_END_INITIALIZATION_CHECK(ext_stream && ext_stream.is_open(), "Cannot open model file.");
-    return &ext_stream;
+    return nullptr;
 }
+OPENVINO_SUPPRESS_DEPRECATED_END
 }  // namespace
 
 FrontEnd::FrontEnd() : m_op_translators(paddle::get_supported_ops()) {}
@@ -392,9 +402,17 @@ bool FrontEnd::supported_impl(const std::vector<ov::Any>& variants) const {
 #endif
     else if (variants[0].is<std::istream*>()) {
         // Validating first stream, it must contain a model
-        auto p_model_stream = variants[0].as<std::istream*>();
+        // step 1:
+        // PDPD API ParseFromIstream always deconstructs the context in model stream.
+        // So, make a copy for variants[0] to avoid breaking the context in variants[0].
+        const auto p_model_stream = variants[0].as<std::istream*>();
+        std::istream copy_model_stream(p_model_stream->rdbuf());
         ::paddle::framework::proto::ProgramDesc fw;
-        return fw.ParseFromIstream(p_model_stream);
+        auto ret = fw.ParseFromIstream(&copy_model_stream);
+        // step 2:
+        // reset the stream position to the beginning.
+        p_model_stream->seekg(0, p_model_stream->beg);
+        return ret;
     }
     return false;
 }
@@ -422,10 +440,10 @@ InputModel::Ptr FrontEnd::load_impl(const std::vector<ov::Any>& variants) const
         }
     } else if (variants.size() == 2 + extra_variants_num) {
         // The case when .pdmodel and .pdparams files are provided
-        std::ifstream model_stream;
-        std::ifstream weights_stream;
-        std::istream* p_model_stream = paddle::variant_to_stream_ptr(variants[0], model_stream);
-        std::istream* p_weights_stream = paddle::variant_to_stream_ptr(variants[1], weights_stream);
+        std::fstream model_fstream, weights_fstream;
+        std::stringstream model_sstream, weights_sstream;
+        std::istream* p_model_stream = paddle::variant_to_stream_ptr(variants[0], model_fstream, model_sstream);
+        std::istream* p_weights_stream = paddle::variant_to_stream_ptr(variants[1], weights_fstream, weights_sstream);
         if (p_model_stream && p_weights_stream) {
             return std::make_shared<InputModel>(std::vector<std::istream*>{p_model_stream, p_weights_stream},
                                                 m_telemetry);
diff --git a/src/frontends/paddle/tests/read_paddle_model_test.cpp b/src/frontends/paddle/tests/read_paddle_model_test.cpp
index 1095c2b8f98110..b362566f52af11 100644
--- a/src/frontends/paddle/tests/read_paddle_model_test.cpp
+++ b/src/frontends/paddle/tests/read_paddle_model_test.cpp
@@ -16,8 +16,62 @@
 #include "frontend/shared/include/utils.hpp"
 #include "openvino/openvino.hpp"
 #include "openvino/opsets/opset1.hpp"
+#include "openvino/opsets/opset8.hpp"
 #include "openvino/pass/serialize.hpp"
 
+TEST(Paddle_Reader_Tests, LoadModelMemoryToCore) {
+    auto model = std::string(TEST_PADDLE_MODELS_DIRNAME) + "conv2d_relu/conv2d_relu.pdmodel";
+    auto param = std::string(TEST_PADDLE_MODELS_DIRNAME) + "conv2d_relu/conv2d_relu.pdiparams";
+
+    ov::Core core;
+    auto read_file = [&](const std::string& file_name, size_t& size) {
+        FILE* sFile = fopen(file_name.c_str(), "r");
+        fseek(sFile, 0, SEEK_END);
+        size = ftell(sFile);
+        uint8_t* ss = (uint8_t*)malloc(size);
+        rewind(sFile);
+        const size_t length = fread(&ss[0], 1, size, sFile);
+        if (size != length) {
+            std::cerr << "file size is not correct\n";
+        }
+        fclose(sFile);
+        return ss;
+    };
+
+    size_t xml_size, bin_size;
+    auto xml_ptr = read_file(model, xml_size);
+    auto bin_ptr = read_file(param, bin_size);
+    ov::Tensor weight_tensor = ov::Tensor(ov::element::u8, {1, bin_size}, bin_ptr);
+    std::string model_str = std::string((char*)xml_ptr, xml_size);
+    auto function = core.read_model(model_str, weight_tensor);
+
+    const auto inputType = ov::element::f32;
+    const auto inputShape = ov::Shape{1, 3, 4, 4};
+    const auto data = std::make_shared<ov::opset1::Parameter>(inputType, inputShape);
+    data->set_friendly_name("xxx");
+    data->output(0).get_tensor().add_names({"xxx"});
+    const auto weight = std::make_shared<ov::opset1::Constant>(ov::element::f32, ov::Shape{5, 3, 1, 1}, 1.0);
+    const auto conv2d = std::make_shared<ov::opset1::Convolution>(data->output(0),
+                                                                  weight->output(0),
+                                                                  ov::Strides({1, 1}),
+                                                                  ov::CoordinateDiff({1, 1}),
+                                                                  ov::CoordinateDiff({1, 1}),
+                                                                  ov::Strides({1, 1}));
+    conv2d->set_friendly_name("conv2d_0.tmp_0");
+    conv2d->output(0).get_tensor().add_names({"conv2d_0.tmp_0"});
+    const auto relu = std::make_shared<ov::opset1::Relu>(conv2d->output(0));
+    relu->set_friendly_name("relu_0.tmp_0");
+    relu->output(0).get_tensor().add_names({"relu_0.tmp_0"});
+    const auto result = std::make_shared<ov::opset1::Result>(relu->output(0));
+    result->set_friendly_name("relu_0.tmp_0/Result");
+    const auto reference = std::make_shared<ov::Model>(ov::NodeVector{result}, ov::ParameterVector{data}, "Model0");
+    const FunctionsComparator func_comparator = FunctionsComparator::with_default().enable(FunctionsComparator::NONE);
+    const FunctionsComparator::Result res = func_comparator(function, reference);
+    ASSERT_TRUE(res.valid) << res.message;
+    free(xml_ptr);
+    free(bin_ptr);
+}
+
 TEST(Paddle_Reader_Tests, ImportBasicModelToCore) {
     auto model = std::string(TEST_PADDLE_MODELS_DIRNAME) + "relu/relu.pdmodel";