Skip to content

The agent engineering platform.

docs: document Anthropic cache TTL count details (#31708)

ccurmeee83993b917d
diff --git a/docs/docs/integrations/chat/anthropic.ipynb b/docs/docs/integrations/chat/anthropic.ipynb
index cfe88b99ff..8a3a20d651 100644
--- a/docs/docs/integrations/chat/anthropic.ipynb
+++ b/docs/docs/integrations/chat/anthropic.ipynb
@@ -568,6 +568,26 @@
" ```\n",
" and specifying `\"cache_control\": {\"type\": \"ephemeral\", \"ttl\": \"1h\"}`.\n",
"\n",
+ " Details of cached token counts will be included on the `InputTokenDetails` of response's `usage_metadata`:\n",
+ "\n",
+ " ```python\n",
+ " response = llm.invoke(messages)\n",
+ " response.usage_metadata\n",
+ " ```\n",
+ " ```\n",
+ " {\n",
+ " \"input_tokens\": 1500,\n",
+ " \"output_tokens\": 200,\n",
+ " \"total_tokens\": 1700,\n",
+ " \"input_token_details\": {\n",
+ " \"cache_read\": 0,\n",
+ " \"cache_creation\": 1000,\n",
+ " \"ephemeral_1h_input_tokens\": 750,\n",
+ " \"ephemeral_5m_input_tokens\": 250,\n",
+ " }\n",
+ " }\n",
+ " ```\n",
+ "\n",
":::"
]
},
diff --git a/libs/core/langchain_core/messages/ai.py b/libs/core/langchain_core/messages/ai.py
index 8641f976a6..c62a03345d 100644
--- a/libs/core/langchain_core/messages/ai.py
+++ b/libs/core/langchain_core/messages/ai.py
@@ -55,6 +55,8 @@ class InputTokenDetails(TypedDict, total=False):
}
.. versionadded:: 0.3.9
+
+ May also hold extra provider-specific keys.
"""
audio: int
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 23bce1d65f..7c07416fe5 100644
--- a/libs/core/tests/unit_tests/prompts/__snapshots__/test_chat.ambr
+++ b/libs/core/tests/unit_tests/prompts/__snapshots__/test_chat.ambr
@@ -702,6 +702,8 @@
}
.. versionadded:: 0.3.9
+
+ May also hold extra provider-specific keys.
''',
'properties': dict({
'audio': dict({
@@ -2132,6 +2134,8 @@
}
.. versionadded:: 0.3.9
+
+ May also hold extra provider-specific keys.
''',
'properties': dict({
'audio': dict({
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 480a6a2b7b..a788c425fc 100644
--- a/libs/core/tests/unit_tests/runnables/__snapshots__/test_graph.ambr
+++ b/libs/core/tests/unit_tests/runnables/__snapshots__/test_graph.ambr
@@ -1105,6 +1105,8 @@
}
.. versionadded:: 0.3.9
+
+ May also hold extra provider-specific keys.
''',
'properties': dict({
'audio': dict({
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 b1519bc3f7..079e490906 100644
--- a/libs/core/tests/unit_tests/runnables/__snapshots__/test_runnable.ambr
+++ b/libs/core/tests/unit_tests/runnables/__snapshots__/test_runnable.ambr
@@ -2650,6 +2650,8 @@
}
.. versionadded:: 0.3.9
+
+ May also hold extra provider-specific keys.
''',
'properties': dict({
'audio': dict({
@@ -4124,6 +4126,8 @@
}
.. versionadded:: 0.3.9
+
+ May also hold extra provider-specific keys.
''',
'properties': dict({
'audio': dict({
@@ -5629,6 +5633,8 @@
}
.. versionadded:: 0.3.9
+
+ May also hold extra provider-specific keys.
''',
'properties': dict({
'audio': dict({
@@ -7009,6 +7015,8 @@
}
.. versionadded:: 0.3.9
+
+ May also hold extra provider-specific keys.
''',
'properties': dict({
'audio': dict({
@@ -8525,6 +8533,8 @@
}
.. versionadded:: 0.3.9
+
+ May also hold extra provider-specific keys.
''',
'properties': dict({
'audio': dict({
@@ -9950,6 +9960,8 @@
}
.. versionadded:: 0.3.9
+
+ May also hold extra provider-specific keys.
''',
'properties': dict({
'audio': dict({
@@ -11374,6 +11386,8 @@
}
.. versionadded:: 0.3.9
+
+ May also hold extra provider-specific keys.
''',
'properties': dict({
'audio': dict({
@@ -12840,6 +12854,8 @@
}
.. versionadded:: 0.3.9
+
+ May also hold extra provider-specific keys.
''',
'properties': dict({
'audio': dict({
diff --git a/libs/partners/anthropic/langchain_anthropic/chat_models.py b/libs/partners/anthropic/langchain_anthropic/chat_models.py
index 581008d3f1..417aeb7f44 100644
--- a/libs/partners/anthropic/langchain_anthropic/chat_models.py
+++ b/libs/partners/anthropic/langchain_anthropic/chat_models.py
@@ -955,6 +955,8 @@ class ChatAnthropic(BaseChatModel):
.. dropdown:: Extended caching
+ .. versionadded:: 0.3.15
+
The cache lifetime is 5 minutes by default. If this is too short, you can
apply one hour caching by enabling the ``"extended-cache-ttl-2025-04-11"``
beta header:
@@ -968,6 +970,28 @@ class ChatAnthropic(BaseChatModel):
and specifying ``"cache_control": {"type": "ephemeral", "ttl": "1h"}``.
+ Details of cached token counts will be included on the ``InputTokenDetails``
+ of response's ``usage_metadata``:
+
+ .. code-block:: python
+
+ response = llm.invoke(messages)
+ response.usage_metadata
+
+ .. code-block:: python
+
+ {
+ "input_tokens": 1500,
+ "output_tokens": 200,
+ "total_tokens": 1700,
+ "input_token_details": {
+ "cache_read": 0,
+ "cache_creation": 1000,
+ "ephemeral_1h_input_tokens": 750,
+ "ephemeral_5m_input_tokens": 250,
+ }
+ }
+
See `Claude documentation <https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching#1-hour-cache-duration-beta>`_
for detail.