diff --git a/libs/core/langchain_core/messages/ai.py b/libs/core/langchain_core/messages/ai.py
index 03a22e79764f5..dece1e575eeb6 100644
--- a/libs/core/langchain_core/messages/ai.py
+++ b/libs/core/langchain_core/messages/ai.py
@@ -2,7 +2,7 @@
 from typing import Any, Literal, Optional, Union
 
 from pydantic import model_validator
-from typing_extensions import Self, TypedDict
+from typing_extensions import NotRequired, Self, TypedDict
 
 from langchain_core.messages.base import (
     BaseMessage,
@@ -29,6 +29,66 @@
 from langchain_core.utils.json import parse_partial_json
 
 
+class InputTokenDetails(TypedDict, total=False):
+    """Breakdown of input token counts.
+
+    Does *not* need to sum to full input token count. Does *not* need to have all keys.
+
+    Example:
+
+        .. code-block:: python
+
+            {
+                "audio": 10,
+                "cache_creation": 200,
+                "cache_read": 100,
+            }
+
+    .. versionadded:: 0.3.9
+    """
+
+    audio: int
+    """Audio input tokens."""
+    cache_creation: int
+    """Input tokens that were cached and there was a cache miss.
+
+    Since there was a cache miss, the cache was created from these tokens.
+    """
+    cache_read: int
+    """Input tokens that were cached and there was a cache hit.
+
+    Since there was a cache hit, the tokens were read from the cache. More precisely,
+    the model state given these tokens was read from the cache.
+    """
+
+
+class OutputTokenDetails(TypedDict, total=False):
+    """Breakdown of output token counts.
+
+    Does *not* need to sum to full output token count. Does *not* need to have all keys.
+
+    Example:
+
+        .. code-block:: python
+
+            {
+                "audio": 10,
+                "reasoning": 200,
+            }
+
+    .. versionadded:: 0.3.9
+    """
+
+    audio: int
+    """Audio output tokens."""
+    reasoning: int
+    """Reasoning output tokens.
+
+    Tokens generated by the model in a chain of thought process (i.e. by OpenAI's o1
+    models) that are not returned as part of model output.
+    """
+
+
 class UsageMetadata(TypedDict):
     """Usage metadata for a message, such as token counts.
 
@@ -39,18 +99,41 @@ class UsageMetadata(TypedDict):
         .. code-block:: python
 
             {
-                "input_tokens": 10,
-                "output_tokens": 20,
-                "total_tokens": 30
+                "input_tokens": 350,
+                "output_tokens": 240,
+                "total_tokens": 590,
+                "input_token_details": {
+                    "audio": 10,
+                    "cache_creation": 200,
+                    "cache_read": 100,
+                },
+                "output_token_details": {
+                    "audio": 10,
+                    "reasoning": 200,
+                }
             }
+
+    .. versionchanged:: 0.3.9
+
+        Added ``input_token_details`` and ``output_token_details``.
     """
 
     input_tokens: int
-    """Count of input (or prompt) tokens."""
+    """Count of input (or prompt) tokens. Sum of all input token types."""
     output_tokens: int
-    """Count of output (or completion) tokens."""
+    """Count of output (or completion) tokens. Sum of all output token types."""
     total_tokens: int
-    """Total token count."""
+    """Total token count. Sum of input_tokens + output_tokens."""
+    input_token_details: NotRequired[InputTokenDetails]
+    """Breakdown of input token counts.
+ 
+    Does *not* need to sum to full input token count. Does *not* need to have all keys.
+    """
+    output_token_details: NotRequired[OutputTokenDetails]
+    """Breakdown of output token counts.
+
+    Does *not* need to sum to full output token count. Does *not* need to have all keys.
+    """
 
 
 class AIMessage(BaseMessage):
diff --git a/libs/core/pyproject.toml b/libs/core/pyproject.toml
index 11a2e85aa376c..20dbf792c952d 100644
--- a/libs/core/pyproject.toml
+++ b/libs/core/pyproject.toml
@@ -45,7 +45,7 @@ python = ">=3.12.4"
 
 [tool.ruff.lint]
 select = [ "B", "C4", "E", "F", "I", "N", "PIE", "SIM", "T201", "UP", "W",]
-ignore = [ "UP007",]
+ignore = [ "UP007", 'W293']
 
 [tool.coverage.run]
 omit = [ "tests/*",]
diff --git a/libs/core/tests/unit_tests/prompts/__snapshots__/test_chat.ambr b/libs/core/tests/unit_tests/prompts/__snapshots__/test_chat.ambr
index 5a235a0f9f933..7e35bd6c46548 100644
--- a/libs/core/tests/unit_tests/prompts/__snapshots__/test_chat.ambr
+++ b/libs/core/tests/unit_tests/prompts/__snapshots__/test_chat.ambr
@@ -677,6 +677,41 @@
         'title': 'HumanMessageChunk',
         'type': 'object',
       }),
+      'InputTokenDetails': dict({
+        'description': '''
+          Breakdown of input token counts.
+          
+          Does *not* need to sum to full input token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "cache_creation": 200,
+                      "cache_read": 100,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'cache_creation': dict({
+            'title': 'Cache Creation',
+            'type': 'integer',
+          }),
+          'cache_read': dict({
+            'title': 'Cache Read',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'InputTokenDetails',
+        'type': 'object',
+      }),
       'InvalidToolCall': dict({
         'description': '''
           Allowance for errors made by LLM.
@@ -743,6 +778,36 @@
         'title': 'InvalidToolCall',
         'type': 'object',
       }),
+      'OutputTokenDetails': dict({
+        'description': '''
+          Breakdown of output token counts.
+          
+          Does *not* need to sum to full output token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "reasoning": 200,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'reasoning': dict({
+            'title': 'Reasoning',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'OutputTokenDetails',
+        'type': 'object',
+      }),
       'SystemMessage': dict({
         'additionalProperties': True,
         'description': '''
@@ -1245,16 +1310,35 @@
               .. code-block:: python
           
                   {
-                      "input_tokens": 10,
-                      "output_tokens": 20,
-                      "total_tokens": 30
+                      "input_tokens": 350,
+                      "output_tokens": 240,
+                      "total_tokens": 590,
+                      "input_token_details": {
+                          "audio": 10,
+                          "cache_creation": 200,
+                          "cache_read": 100,
+                      },
+                      "output_token_details": {
+                          "audio": 10,
+                          "reasoning": 200,
+                      }
                   }
+          
+          .. versionchanged:: 0.3.9
+          
+              Added ``input_token_details`` and ``output_token_details``.
         ''',
         'properties': dict({
+          'input_token_details': dict({
+            '$ref': '#/$defs/InputTokenDetails',
+          }),
           'input_tokens': dict({
             'title': 'Input Tokens',
             'type': 'integer',
           }),
+          'output_token_details': dict({
+            '$ref': '#/$defs/OutputTokenDetails',
+          }),
           'output_tokens': dict({
             'title': 'Output Tokens',
             'type': 'integer',
@@ -2008,6 +2092,41 @@
         'title': 'HumanMessageChunk',
         'type': 'object',
       }),
+      'InputTokenDetails': dict({
+        'description': '''
+          Breakdown of input token counts.
+          
+          Does *not* need to sum to full input token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "cache_creation": 200,
+                      "cache_read": 100,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'cache_creation': dict({
+            'title': 'Cache Creation',
+            'type': 'integer',
+          }),
+          'cache_read': dict({
+            'title': 'Cache Read',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'InputTokenDetails',
+        'type': 'object',
+      }),
       'InvalidToolCall': dict({
         'description': '''
           Allowance for errors made by LLM.
@@ -2074,6 +2193,36 @@
         'title': 'InvalidToolCall',
         'type': 'object',
       }),
+      'OutputTokenDetails': dict({
+        'description': '''
+          Breakdown of output token counts.
+          
+          Does *not* need to sum to full output token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "reasoning": 200,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'reasoning': dict({
+            'title': 'Reasoning',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'OutputTokenDetails',
+        'type': 'object',
+      }),
       'SystemMessage': dict({
         'additionalProperties': True,
         'description': '''
@@ -2576,16 +2725,35 @@
               .. code-block:: python
           
                   {
-                      "input_tokens": 10,
-                      "output_tokens": 20,
-                      "total_tokens": 30
+                      "input_tokens": 350,
+                      "output_tokens": 240,
+                      "total_tokens": 590,
+                      "input_token_details": {
+                          "audio": 10,
+                          "cache_creation": 200,
+                          "cache_read": 100,
+                      },
+                      "output_token_details": {
+                          "audio": 10,
+                          "reasoning": 200,
+                      }
                   }
+          
+          .. versionchanged:: 0.3.9
+          
+              Added ``input_token_details`` and ``output_token_details``.
         ''',
         'properties': dict({
+          'input_token_details': dict({
+            '$ref': '#/$defs/InputTokenDetails',
+          }),
           'input_tokens': dict({
             'title': 'Input Tokens',
             'type': 'integer',
           }),
+          'output_token_details': dict({
+            '$ref': '#/$defs/OutputTokenDetails',
+          }),
           'output_tokens': dict({
             'title': 'Output Tokens',
             'type': 'integer',
diff --git a/libs/core/tests/unit_tests/runnables/__snapshots__/test_graph.ambr b/libs/core/tests/unit_tests/runnables/__snapshots__/test_graph.ambr
index aa70b9bb5a3e9..ba9d742b37407 100644
--- a/libs/core/tests/unit_tests/runnables/__snapshots__/test_graph.ambr
+++ b/libs/core/tests/unit_tests/runnables/__snapshots__/test_graph.ambr
@@ -1037,6 +1037,41 @@
               'title': 'HumanMessageChunk',
               'type': 'object',
             }),
+            'InputTokenDetails': dict({
+              'description': '''
+                Breakdown of input token counts.
+                
+                Does *not* need to sum to full input token count. Does *not* need to have all keys.
+                
+                Example:
+                
+                    .. code-block:: python
+                
+                        {
+                            "audio": 10,
+                            "cache_creation": 200,
+                            "cache_read": 100,
+                        }
+                
+                .. versionadded:: 0.3.9
+              ''',
+              'properties': dict({
+                'audio': dict({
+                  'title': 'Audio',
+                  'type': 'integer',
+                }),
+                'cache_creation': dict({
+                  'title': 'Cache Creation',
+                  'type': 'integer',
+                }),
+                'cache_read': dict({
+                  'title': 'Cache Read',
+                  'type': 'integer',
+                }),
+              }),
+              'title': 'InputTokenDetails',
+              'type': 'object',
+            }),
             'InvalidToolCall': dict({
               'description': '''
                 Allowance for errors made by LLM.
@@ -1103,6 +1138,36 @@
               'title': 'InvalidToolCall',
               'type': 'object',
             }),
+            'OutputTokenDetails': dict({
+              'description': '''
+                Breakdown of output token counts.
+                
+                Does *not* need to sum to full output token count. Does *not* need to have all keys.
+                
+                Example:
+                
+                    .. code-block:: python
+                
+                        {
+                            "audio": 10,
+                            "reasoning": 200,
+                        }
+                
+                .. versionadded:: 0.3.9
+              ''',
+              'properties': dict({
+                'audio': dict({
+                  'title': 'Audio',
+                  'type': 'integer',
+                }),
+                'reasoning': dict({
+                  'title': 'Reasoning',
+                  'type': 'integer',
+                }),
+              }),
+              'title': 'OutputTokenDetails',
+              'type': 'object',
+            }),
             'SystemMessage': dict({
               'additionalProperties': True,
               'description': '''
@@ -1605,16 +1670,35 @@
                     .. code-block:: python
                 
                         {
-                            "input_tokens": 10,
-                            "output_tokens": 20,
-                            "total_tokens": 30
+                            "input_tokens": 350,
+                            "output_tokens": 240,
+                            "total_tokens": 590,
+                            "input_token_details": {
+                                "audio": 10,
+                                "cache_creation": 200,
+                                "cache_read": 100,
+                            },
+                            "output_token_details": {
+                                "audio": 10,
+                                "reasoning": 200,
+                            }
                         }
+                
+                .. versionchanged:: 0.3.9
+                
+                    Added ``input_token_details`` and ``output_token_details``.
               ''',
               'properties': dict({
+                'input_token_details': dict({
+                  '$ref': '#/$defs/InputTokenDetails',
+                }),
                 'input_tokens': dict({
                   'title': 'Input Tokens',
                   'type': 'integer',
                 }),
+                'output_token_details': dict({
+                  '$ref': '#/$defs/OutputTokenDetails',
+                }),
                 'output_tokens': dict({
                   'title': 'Output Tokens',
                   'type': 'integer',
diff --git a/libs/core/tests/unit_tests/runnables/__snapshots__/test_runnable.ambr b/libs/core/tests/unit_tests/runnables/__snapshots__/test_runnable.ambr
index 045b9eced274c..85b53552716a2 100644
--- a/libs/core/tests/unit_tests/runnables/__snapshots__/test_runnable.ambr
+++ b/libs/core/tests/unit_tests/runnables/__snapshots__/test_runnable.ambr
@@ -2639,6 +2639,41 @@
         'title': 'HumanMessageChunk',
         'type': 'object',
       }),
+      'InputTokenDetails': dict({
+        'description': '''
+          Breakdown of input token counts.
+          
+          Does *not* need to sum to full input token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "cache_creation": 200,
+                      "cache_read": 100,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'cache_creation': dict({
+            'title': 'Cache Creation',
+            'type': 'integer',
+          }),
+          'cache_read': dict({
+            'title': 'Cache Read',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'InputTokenDetails',
+        'type': 'object',
+      }),
       'InvalidToolCall': dict({
         'description': '''
           Allowance for errors made by LLM.
@@ -2705,6 +2740,36 @@
         'title': 'InvalidToolCall',
         'type': 'object',
       }),
+      'OutputTokenDetails': dict({
+        'description': '''
+          Breakdown of output token counts.
+          
+          Does *not* need to sum to full output token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "reasoning": 200,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'reasoning': dict({
+            'title': 'Reasoning',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'OutputTokenDetails',
+        'type': 'object',
+      }),
       'SystemMessage': dict({
         'additionalProperties': True,
         'description': '''
@@ -3207,16 +3272,35 @@
               .. code-block:: python
           
                   {
-                      "input_tokens": 10,
-                      "output_tokens": 20,
-                      "total_tokens": 30
+                      "input_tokens": 350,
+                      "output_tokens": 240,
+                      "total_tokens": 590,
+                      "input_token_details": {
+                          "audio": 10,
+                          "cache_creation": 200,
+                          "cache_read": 100,
+                      },
+                      "output_token_details": {
+                          "audio": 10,
+                          "reasoning": 200,
+                      }
                   }
+          
+          .. versionchanged:: 0.3.9
+          
+              Added ``input_token_details`` and ``output_token_details``.
         ''',
         'properties': dict({
+          'input_token_details': dict({
+            '$ref': '#/$defs/InputTokenDetails',
+          }),
           'input_tokens': dict({
             'title': 'Input Tokens',
             'type': 'integer',
           }),
+          'output_token_details': dict({
+            '$ref': '#/$defs/OutputTokenDetails',
+          }),
           'output_tokens': dict({
             'title': 'Output Tokens',
             'type': 'integer',
@@ -4028,6 +4112,41 @@
         'title': 'HumanMessageChunk',
         'type': 'object',
       }),
+      'InputTokenDetails': dict({
+        'description': '''
+          Breakdown of input token counts.
+          
+          Does *not* need to sum to full input token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "cache_creation": 200,
+                      "cache_read": 100,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'cache_creation': dict({
+            'title': 'Cache Creation',
+            'type': 'integer',
+          }),
+          'cache_read': dict({
+            'title': 'Cache Read',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'InputTokenDetails',
+        'type': 'object',
+      }),
       'InvalidToolCall': dict({
         'description': '''
           Allowance for errors made by LLM.
@@ -4094,6 +4213,36 @@
         'title': 'InvalidToolCall',
         'type': 'object',
       }),
+      'OutputTokenDetails': dict({
+        'description': '''
+          Breakdown of output token counts.
+          
+          Does *not* need to sum to full output token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "reasoning": 200,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'reasoning': dict({
+            'title': 'Reasoning',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'OutputTokenDetails',
+        'type': 'object',
+      }),
       'StringPromptValue': dict({
         'description': 'String prompt value.',
         'properties': dict({
@@ -4615,16 +4764,35 @@
               .. code-block:: python
           
                   {
-                      "input_tokens": 10,
-                      "output_tokens": 20,
-                      "total_tokens": 30
+                      "input_tokens": 350,
+                      "output_tokens": 240,
+                      "total_tokens": 590,
+                      "input_token_details": {
+                          "audio": 10,
+                          "cache_creation": 200,
+                          "cache_read": 100,
+                      },
+                      "output_token_details": {
+                          "audio": 10,
+                          "reasoning": 200,
+                      }
                   }
+          
+          .. versionchanged:: 0.3.9
+          
+              Added ``input_token_details`` and ``output_token_details``.
         ''',
         'properties': dict({
+          'input_token_details': dict({
+            '$ref': '#/$defs/InputTokenDetails',
+          }),
           'input_tokens': dict({
             'title': 'Input Tokens',
             'type': 'integer',
           }),
+          'output_token_details': dict({
+            '$ref': '#/$defs/OutputTokenDetails',
+          }),
           'output_tokens': dict({
             'title': 'Output Tokens',
             'type': 'integer',
@@ -5448,6 +5616,41 @@
         'title': 'HumanMessageChunk',
         'type': 'object',
       }),
+      'InputTokenDetails': dict({
+        'description': '''
+          Breakdown of input token counts.
+          
+          Does *not* need to sum to full input token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "cache_creation": 200,
+                      "cache_read": 100,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'cache_creation': dict({
+            'title': 'Cache Creation',
+            'type': 'integer',
+          }),
+          'cache_read': dict({
+            'title': 'Cache Read',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'InputTokenDetails',
+        'type': 'object',
+      }),
       'InvalidToolCall': dict({
         'description': '''
           Allowance for errors made by LLM.
@@ -5514,6 +5717,36 @@
         'title': 'InvalidToolCall',
         'type': 'object',
       }),
+      'OutputTokenDetails': dict({
+        'description': '''
+          Breakdown of output token counts.
+          
+          Does *not* need to sum to full output token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "reasoning": 200,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'reasoning': dict({
+            'title': 'Reasoning',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'OutputTokenDetails',
+        'type': 'object',
+      }),
       'StringPromptValue': dict({
         'description': 'String prompt value.',
         'properties': dict({
@@ -6035,16 +6268,35 @@
               .. code-block:: python
           
                   {
-                      "input_tokens": 10,
-                      "output_tokens": 20,
-                      "total_tokens": 30
+                      "input_tokens": 350,
+                      "output_tokens": 240,
+                      "total_tokens": 590,
+                      "input_token_details": {
+                          "audio": 10,
+                          "cache_creation": 200,
+                          "cache_read": 100,
+                      },
+                      "output_token_details": {
+                          "audio": 10,
+                          "reasoning": 200,
+                      }
                   }
+          
+          .. versionchanged:: 0.3.9
+          
+              Added ``input_token_details`` and ``output_token_details``.
         ''',
         'properties': dict({
+          'input_token_details': dict({
+            '$ref': '#/definitions/InputTokenDetails',
+          }),
           'input_tokens': dict({
             'title': 'Input Tokens',
             'type': 'integer',
           }),
+          'output_token_details': dict({
+            '$ref': '#/definitions/OutputTokenDetails',
+          }),
           'output_tokens': dict({
             'title': 'Output Tokens',
             'type': 'integer',
@@ -6744,6 +6996,41 @@
         'title': 'HumanMessageChunk',
         'type': 'object',
       }),
+      'InputTokenDetails': dict({
+        'description': '''
+          Breakdown of input token counts.
+          
+          Does *not* need to sum to full input token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "cache_creation": 200,
+                      "cache_read": 100,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'cache_creation': dict({
+            'title': 'Cache Creation',
+            'type': 'integer',
+          }),
+          'cache_read': dict({
+            'title': 'Cache Read',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'InputTokenDetails',
+        'type': 'object',
+      }),
       'InvalidToolCall': dict({
         'description': '''
           Allowance for errors made by LLM.
@@ -6810,6 +7097,36 @@
         'title': 'InvalidToolCall',
         'type': 'object',
       }),
+      'OutputTokenDetails': dict({
+        'description': '''
+          Breakdown of output token counts.
+          
+          Does *not* need to sum to full output token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "reasoning": 200,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'reasoning': dict({
+            'title': 'Reasoning',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'OutputTokenDetails',
+        'type': 'object',
+      }),
       'SystemMessage': dict({
         'additionalProperties': True,
         'description': '''
@@ -7312,16 +7629,35 @@
               .. code-block:: python
           
                   {
-                      "input_tokens": 10,
-                      "output_tokens": 20,
-                      "total_tokens": 30
+                      "input_tokens": 350,
+                      "output_tokens": 240,
+                      "total_tokens": 590,
+                      "input_token_details": {
+                          "audio": 10,
+                          "cache_creation": 200,
+                          "cache_read": 100,
+                      },
+                      "output_token_details": {
+                          "audio": 10,
+                          "reasoning": 200,
+                      }
                   }
+          
+          .. versionchanged:: 0.3.9
+          
+              Added ``input_token_details`` and ``output_token_details``.
         ''',
         'properties': dict({
+          'input_token_details': dict({
+            '$ref': '#/definitions/InputTokenDetails',
+          }),
           'input_tokens': dict({
             'title': 'Input Tokens',
             'type': 'integer',
           }),
+          'output_token_details': dict({
+            '$ref': '#/definitions/OutputTokenDetails',
+          }),
           'output_tokens': dict({
             'title': 'Output Tokens',
             'type': 'integer',
@@ -8175,6 +8511,41 @@
         'title': 'HumanMessageChunk',
         'type': 'object',
       }),
+      'InputTokenDetails': dict({
+        'description': '''
+          Breakdown of input token counts.
+          
+          Does *not* need to sum to full input token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "cache_creation": 200,
+                      "cache_read": 100,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'cache_creation': dict({
+            'title': 'Cache Creation',
+            'type': 'integer',
+          }),
+          'cache_read': dict({
+            'title': 'Cache Read',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'InputTokenDetails',
+        'type': 'object',
+      }),
       'InvalidToolCall': dict({
         'description': '''
           Allowance for errors made by LLM.
@@ -8241,6 +8612,36 @@
         'title': 'InvalidToolCall',
         'type': 'object',
       }),
+      'OutputTokenDetails': dict({
+        'description': '''
+          Breakdown of output token counts.
+          
+          Does *not* need to sum to full output token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "reasoning": 200,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'reasoning': dict({
+            'title': 'Reasoning',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'OutputTokenDetails',
+        'type': 'object',
+      }),
       'StringPromptValue': dict({
         'description': 'String prompt value.',
         'properties': dict({
@@ -8762,16 +9163,35 @@
               .. code-block:: python
           
                   {
-                      "input_tokens": 10,
-                      "output_tokens": 20,
-                      "total_tokens": 30
+                      "input_tokens": 350,
+                      "output_tokens": 240,
+                      "total_tokens": 590,
+                      "input_token_details": {
+                          "audio": 10,
+                          "cache_creation": 200,
+                          "cache_read": 100,
+                      },
+                      "output_token_details": {
+                          "audio": 10,
+                          "reasoning": 200,
+                      }
                   }
+          
+          .. versionchanged:: 0.3.9
+          
+              Added ``input_token_details`` and ``output_token_details``.
         ''',
         'properties': dict({
+          'input_token_details': dict({
+            '$ref': '#/definitions/InputTokenDetails',
+          }),
           'input_tokens': dict({
             'title': 'Input Tokens',
             'type': 'integer',
           }),
+          'output_token_details': dict({
+            '$ref': '#/definitions/OutputTokenDetails',
+          }),
           'output_tokens': dict({
             'title': 'Output Tokens',
             'type': 'integer',
@@ -9516,6 +9936,41 @@
         'title': 'HumanMessageChunk',
         'type': 'object',
       }),
+      'InputTokenDetails': dict({
+        'description': '''
+          Breakdown of input token counts.
+          
+          Does *not* need to sum to full input token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "cache_creation": 200,
+                      "cache_read": 100,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'cache_creation': dict({
+            'title': 'Cache Creation',
+            'type': 'integer',
+          }),
+          'cache_read': dict({
+            'title': 'Cache Read',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'InputTokenDetails',
+        'type': 'object',
+      }),
       'InvalidToolCall': dict({
         'description': '''
           Allowance for errors made by LLM.
@@ -9582,6 +10037,36 @@
         'title': 'InvalidToolCall',
         'type': 'object',
       }),
+      'OutputTokenDetails': dict({
+        'description': '''
+          Breakdown of output token counts.
+          
+          Does *not* need to sum to full output token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "reasoning": 200,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'reasoning': dict({
+            'title': 'Reasoning',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'OutputTokenDetails',
+        'type': 'object',
+      }),
       'SystemMessage': dict({
         'additionalProperties': True,
         'description': '''
@@ -10084,16 +10569,35 @@
               .. code-block:: python
           
                   {
-                      "input_tokens": 10,
-                      "output_tokens": 20,
-                      "total_tokens": 30
+                      "input_tokens": 350,
+                      "output_tokens": 240,
+                      "total_tokens": 590,
+                      "input_token_details": {
+                          "audio": 10,
+                          "cache_creation": 200,
+                          "cache_read": 100,
+                      },
+                      "output_token_details": {
+                          "audio": 10,
+                          "reasoning": 200,
+                      }
                   }
+          
+          .. versionchanged:: 0.3.9
+          
+              Added ``input_token_details`` and ``output_token_details``.
         ''',
         'properties': dict({
+          'input_token_details': dict({
+            '$ref': '#/definitions/InputTokenDetails',
+          }),
           'input_tokens': dict({
             'title': 'Input Tokens',
             'type': 'integer',
           }),
+          'output_token_details': dict({
+            '$ref': '#/definitions/OutputTokenDetails',
+          }),
           'output_tokens': dict({
             'title': 'Output Tokens',
             'type': 'integer',
@@ -10855,6 +11359,41 @@
         'title': 'HumanMessageChunk',
         'type': 'object',
       }),
+      'InputTokenDetails': dict({
+        'description': '''
+          Breakdown of input token counts.
+          
+          Does *not* need to sum to full input token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "cache_creation": 200,
+                      "cache_read": 100,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'cache_creation': dict({
+            'title': 'Cache Creation',
+            'type': 'integer',
+          }),
+          'cache_read': dict({
+            'title': 'Cache Read',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'InputTokenDetails',
+        'type': 'object',
+      }),
       'InvalidToolCall': dict({
         'description': '''
           Allowance for errors made by LLM.
@@ -10921,6 +11460,36 @@
         'title': 'InvalidToolCall',
         'type': 'object',
       }),
+      'OutputTokenDetails': dict({
+        'description': '''
+          Breakdown of output token counts.
+          
+          Does *not* need to sum to full output token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "reasoning": 200,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'reasoning': dict({
+            'title': 'Reasoning',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'OutputTokenDetails',
+        'type': 'object',
+      }),
       'PromptTemplateOutput': dict({
         'anyOf': list([
           dict({
@@ -11453,16 +12022,35 @@
               .. code-block:: python
           
                   {
-                      "input_tokens": 10,
-                      "output_tokens": 20,
-                      "total_tokens": 30
+                      "input_tokens": 350,
+                      "output_tokens": 240,
+                      "total_tokens": 590,
+                      "input_token_details": {
+                          "audio": 10,
+                          "cache_creation": 200,
+                          "cache_read": 100,
+                      },
+                      "output_token_details": {
+                          "audio": 10,
+                          "reasoning": 200,
+                      }
                   }
+          
+          .. versionchanged:: 0.3.9
+          
+              Added ``input_token_details`` and ``output_token_details``.
         ''',
         'properties': dict({
+          'input_token_details': dict({
+            '$ref': '#/definitions/InputTokenDetails',
+          }),
           'input_tokens': dict({
             'title': 'Input Tokens',
             'type': 'integer',
           }),
+          'output_token_details': dict({
+            '$ref': '#/definitions/OutputTokenDetails',
+          }),
           'output_tokens': dict({
             'title': 'Output Tokens',
             'type': 'integer',
@@ -12236,6 +12824,41 @@
         'title': 'HumanMessageChunk',
         'type': 'object',
       }),
+      'InputTokenDetails': dict({
+        'description': '''
+          Breakdown of input token counts.
+          
+          Does *not* need to sum to full input token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "cache_creation": 200,
+                      "cache_read": 100,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'cache_creation': dict({
+            'title': 'Cache Creation',
+            'type': 'integer',
+          }),
+          'cache_read': dict({
+            'title': 'Cache Read',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'InputTokenDetails',
+        'type': 'object',
+      }),
       'InvalidToolCall': dict({
         'description': '''
           Allowance for errors made by LLM.
@@ -12302,6 +12925,36 @@
         'title': 'InvalidToolCall',
         'type': 'object',
       }),
+      'OutputTokenDetails': dict({
+        'description': '''
+          Breakdown of output token counts.
+          
+          Does *not* need to sum to full output token count. Does *not* need to have all keys.
+          
+          Example:
+          
+              .. code-block:: python
+          
+                  {
+                      "audio": 10,
+                      "reasoning": 200,
+                  }
+          
+          .. versionadded:: 0.3.9
+        ''',
+        'properties': dict({
+          'audio': dict({
+            'title': 'Audio',
+            'type': 'integer',
+          }),
+          'reasoning': dict({
+            'title': 'Reasoning',
+            'type': 'integer',
+          }),
+        }),
+        'title': 'OutputTokenDetails',
+        'type': 'object',
+      }),
       'StringPromptValue': dict({
         'description': 'String prompt value.',
         'properties': dict({
@@ -12823,16 +13476,35 @@
               .. code-block:: python
           
                   {
-                      "input_tokens": 10,
-                      "output_tokens": 20,
-                      "total_tokens": 30
+                      "input_tokens": 350,
+                      "output_tokens": 240,
+                      "total_tokens": 590,
+                      "input_token_details": {
+                          "audio": 10,
+                          "cache_creation": 200,
+                          "cache_read": 100,
+                      },
+                      "output_token_details": {
+                          "audio": 10,
+                          "reasoning": 200,
+                      }
                   }
+          
+          .. versionchanged:: 0.3.9
+          
+              Added ``input_token_details`` and ``output_token_details``.
         ''',
         'properties': dict({
+          'input_token_details': dict({
+            '$ref': '#/definitions/InputTokenDetails',
+          }),
           'input_tokens': dict({
             'title': 'Input Tokens',
             'type': 'integer',
           }),
+          'output_token_details': dict({
+            '$ref': '#/definitions/OutputTokenDetails',
+          }),
           'output_tokens': dict({
             'title': 'Output Tokens',
             'type': 'integer',
diff --git a/libs/standard-tests/langchain_standard_tests/integration_tests/chat_models.py b/libs/standard-tests/langchain_standard_tests/integration_tests/chat_models.py
index c5ac855cdc6a2..1da209095ef40 100644
--- a/libs/standard-tests/langchain_standard_tests/integration_tests/chat_models.py
+++ b/libs/standard-tests/langchain_standard_tests/integration_tests/chat_models.py
@@ -151,6 +151,31 @@ def test_usage_metadata(self, model: BaseChatModel) -> None:
         assert isinstance(result.usage_metadata["output_tokens"], int)
         assert isinstance(result.usage_metadata["total_tokens"], int)
 
+        if "audio_input" in self.supported_usage_metadata_details:
+            msg = self.invoke_with_audio_input()
+            assert isinstance(msg.usage_metadata["input_token_details"]["audio"], int)  # type: ignore[index]
+        if "audio_output" in self.supported_usage_metadata_details:
+            msg = self.invoke_with_audio_output()
+            assert isinstance(msg.usage_metadata["output_token_details"]["audio"], int)  # type: ignore[index]
+        if "reasoning_output" in self.supported_usage_metadata_details:
+            msg = self.invoke_with_reasoning_output()
+            assert isinstance(
+                msg.usage_metadata["output_token_details"]["reasoning"],  # type: ignore[index]
+                int,
+            )
+        if "cache_read_input" in self.supported_usage_metadata_details:
+            msg = self.invoke_with_cache_read_input()
+            assert isinstance(
+                msg.usage_metadata["input_token_details"]["cache_read"],  # type: ignore[index]
+                int,
+            )
+        if "cache_creation_input" in self.supported_usage_metadata_details:
+            msg = self.invoke_with_cache_creation_input()
+            assert isinstance(
+                msg.usage_metadata["input_token_details"]["cache_creation"],  # type: ignore[index]
+                int,
+            )
+
     def test_usage_metadata_streaming(self, model: BaseChatModel) -> None:
         if not self.returns_usage_metadata:
             pytest.skip("Not implemented.")
@@ -164,6 +189,31 @@ def test_usage_metadata_streaming(self, model: BaseChatModel) -> None:
         assert isinstance(full.usage_metadata["output_tokens"], int)
         assert isinstance(full.usage_metadata["total_tokens"], int)
 
+        if "audio_input" in self.supported_usage_metadata_details:
+            msg = self.invoke_with_audio_input(stream=True)
+            assert isinstance(msg.usage_metadata["input_token_details"]["audio"], int)  # type: ignore[index]
+        if "audio_output" in self.supported_usage_metadata_details:
+            msg = self.invoke_with_audio_output(stream=True)
+            assert isinstance(msg.usage_metadata["output_token_details"]["audio"], int)  # type: ignore[index]
+        if "reasoning_output" in self.supported_usage_metadata_details:
+            msg = self.invoke_with_reasoning_output(stream=True)
+            assert isinstance(
+                msg.usage_metadata["output_token_details"]["reasoning"],  # type: ignore[index]
+                int,
+            )
+        if "cache_read_input" in self.supported_usage_metadata_details:
+            msg = self.invoke_with_cache_read_input(stream=True)
+            assert isinstance(
+                msg.usage_metadata["input_token_details"]["cache_read"],  # type: ignore[index]
+                int,
+            )
+        if "cache_creation_input" in self.supported_usage_metadata_details:
+            msg = self.invoke_with_cache_creation_input(stream=True)
+            assert isinstance(
+                msg.usage_metadata["input_token_details"]["cache_creation"],  # type: ignore[index]
+                int,
+            )
+
     def test_stop_sequence(self, model: BaseChatModel) -> None:
         result = model.invoke("hi", stop=["you"])
         assert isinstance(result, AIMessage)
@@ -608,3 +658,18 @@ def test_message_with_name(self, model: BaseChatModel) -> None:
         assert isinstance(result, AIMessage)
         assert isinstance(result.content, str)
         assert len(result.content) > 0
+
+    def invoke_with_audio_input(self, *, stream: bool = False) -> AIMessage:
+        raise NotImplementedError()
+
+    def invoke_with_audio_output(self, *, stream: bool = False) -> AIMessage:
+        raise NotImplementedError()
+
+    def invoke_with_reasoning_output(self, *, stream: bool = False) -> AIMessage:
+        raise NotImplementedError()
+
+    def invoke_with_cache_read_input(self, *, stream: bool = False) -> AIMessage:
+        raise NotImplementedError()
+
+    def invoke_with_cache_creation_input(self, *, stream: bool = False) -> AIMessage:
+        raise NotImplementedError()
diff --git a/libs/standard-tests/langchain_standard_tests/unit_tests/chat_models.py b/libs/standard-tests/langchain_standard_tests/unit_tests/chat_models.py
index 1a611f1800f5c..1298c022852cf 100644
--- a/libs/standard-tests/langchain_standard_tests/unit_tests/chat_models.py
+++ b/libs/standard-tests/langchain_standard_tests/unit_tests/chat_models.py
@@ -138,6 +138,20 @@ def supports_anthropic_inputs(self) -> bool:
     def supports_image_tool_message(self) -> bool:
         return False
 
+    @property
+    def supported_usage_metadata_details(
+        self,
+    ) -> List[
+        Literal[
+            "audio_input",
+            "audio_output",
+            "reasoning_output",
+            "cache_read_input",
+            "cache_creation_input",
+        ]
+    ]:
+        return []
+
 
 class ChatModelUnitTests(ChatModelTests):
     @property