diff --git a/tools/accuracy_checker/accuracy_checker/adapters/README.md b/tools/accuracy_checker/accuracy_checker/adapters/README.md
index 26b40d1df9..d7eae48755 100644
--- a/tools/accuracy_checker/accuracy_checker/adapters/README.md
+++ b/tools/accuracy_checker/accuracy_checker/adapters/README.md
@@ -43,6 +43,7 @@ AccuracyChecker supports following set of adapters:
   * `joining_method` - method used to join embeddings (optional, supported methods are `sum` and `concatenation`, default - `sum`).
   * `target_out` - target output layer name (Optional, if not provided first in the model will be used).
   * `keep_shape` - allow keeping initial shape for predicted embedding (Optional, default `False`, it means that model output will be flattenized).
+  * `mean_pooling` - average the embeddings of all tokens from last_hidden_state (Optional, default `False`)
 * `yolo_v2` - converting output of YOLO v2 family models to `DetectionPrediction` representation.
   * `classes` - number of detection classes (default 20).
   * `anchors` - anchor values provided as comma-separated list or one of precomputed:
diff --git a/tools/accuracy_checker/accuracy_checker/adapters/reidentification.py b/tools/accuracy_checker/accuracy_checker/adapters/reidentification.py
index e2c4618c45..dea6a59799 100644
--- a/tools/accuracy_checker/accuracy_checker/adapters/reidentification.py
+++ b/tools/accuracy_checker/accuracy_checker/adapters/reidentification.py
@@ -1,5 +1,5 @@
 """
-Copyright (c) 2018-2024 Intel Corporation
+Copyright (c) 2018-2025 Intel Corporation
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -41,7 +41,9 @@ def parameters(cls):
                 choices=['sum', 'concatenation']
             ),
             'target_out': StringField(optional=True, description='Target output layer name'),
-            'keep_shape': BoolField(optional=True, default=False, description='keep output embedding shape')
+            'keep_shape': BoolField(optional=True, default=False, description='keep output embedding shape'),
+            'mean_pooling': BoolField(optional=True, default=False,
+                                      description='Average the embeddings of all tokens for last_hidden_state')
         })
 
         return parameters
@@ -54,6 +56,7 @@ def configure(self):
         self.joining_method = self.get_value_from_config('joining_method')
         self.target_out = self.get_value_from_config('target_out')
         self.keep_shape = self.get_value_from_config('keep_shape')
+        self.mean_pooling = self.get_value_from_config('mean_pooling')
 
     def process(self, raw, identifiers, frame_meta):
         """
@@ -67,6 +70,10 @@ def process(self, raw, identifiers, frame_meta):
         raw_prediction = self._extract_predictions(raw, frame_meta)
         prediction = raw_prediction[self.output_blob]
 
+        if self.mean_pooling:
+            # Shape: (1, 128, 768) -> (1, 768)
+            prediction = np.mean(prediction, axis=1)
+
         if self.grn_workaround:
             # workaround: GRN layer
             prediction = self._grn_layer(prediction)