From db20c23adc43911b6560c5e9e95299ae7b1cff3b Mon Sep 17 00:00:00 2001
From: James Hargreaves <james.hargreaves@portswigger.net>
Date: Wed, 17 Jun 2026 16:47:24 +0100
Subject: [PATCH] feat(posthog): surface Anthropic cache token counts in
 StandardLoggingPayload and $ai_generation events

---
 litellm/integrations/posthog.py               |  6 ++++++
 litellm/litellm_core_utils/litellm_logging.py | 12 ++++++++++--
 litellm/types/utils.py                        |  2 ++
 3 files changed, 18 insertions(+), 2 deletions(-)

diff --git a/litellm/integrations/posthog.py b/litellm/integrations/posthog.py
index 072ae4945a0..1651e3b5ba7 100644
--- a/litellm/integrations/posthog.py
+++ b/litellm/integrations/posthog.py
@@ -229,6 +229,12 @@ class PostHogLogger(CustomBatchLogger):
             properties["$ai_output_tokens"] = self._safe_get(
                 standard_logging_object, "completion_tokens", 0
             )
+        cache_read = self._safe_get(standard_logging_object, "cache_read_input_tokens", 0) or 0
+        cache_creation = self._safe_get(standard_logging_object, "cache_creation_input_tokens", 0) or 0
+        if cache_read:
+            properties["$ai_cache_read_input_tokens"] = cache_read
+        if cache_creation:
+            properties["$ai_cache_creation_input_tokens"] = cache_creation
 
         # Cost and performance
         response_cost = self._safe_get(standard_logging_object, "response_cost")
diff --git a/litellm/litellm_core_utils/litellm_logging.py b/litellm/litellm_core_utils/litellm_logging.py
index d750a509054..1a79bd2b753 100644
--- a/litellm/litellm_core_utils/litellm_logging.py
+++ b/litellm/litellm_core_utils/litellm_logging.py
@@ -5174,7 +5174,10 @@ class StandardLoggingPayloadSetup:
         """
         _empty: dict = {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0}
         if combined_usage_object is not None:
-            return combined_usage_object.model_dump()
+            d = combined_usage_object.model_dump()
+            d["cache_read_input_tokens"] = combined_usage_object._cache_read_input_tokens
+            d["cache_creation_input_tokens"] = combined_usage_object._cache_creation_input_tokens
+            return d
         if not response_obj:
             return _empty
         _raw = response_obj.get("usage", None)
@@ -5193,7 +5196,10 @@ class StandardLoggingPayloadSetup:
                 )
             return _raw
         if isinstance(_raw, Usage):
-            return _raw.model_dump()
+            d = _raw.model_dump()
+            d["cache_read_input_tokens"] = _raw._cache_read_input_tokens
+            d["cache_creation_input_tokens"] = _raw._cache_creation_input_tokens
+            return d
         return _empty
 
     @staticmethod
@@ -5890,6 +5896,8 @@ def get_standard_logging_object_payload(
             total_tokens=usage_dict.get("total_tokens", 0),
             prompt_tokens=usage_dict.get("prompt_tokens", 0),
             completion_tokens=usage_dict.get("completion_tokens", 0),
+            cache_read_input_tokens=usage_dict.get("cache_read_input_tokens", 0),
+            cache_creation_input_tokens=usage_dict.get("cache_creation_input_tokens", 0),
             request_tags=request_tags,
             end_user=end_user_id or "",
             api_base=StandardLoggingPayloadSetup.strip_trailing_slash(
diff --git a/litellm/types/utils.py b/litellm/types/utils.py
index 32bfc8835fe..b83fb9b38e6 100644
--- a/litellm/types/utils.py
+++ b/litellm/types/utils.py
@@ -2979,6 +2979,8 @@ class StandardLoggingPayload(TypedDict):
     total_tokens: int
     prompt_tokens: int
     completion_tokens: int
+    cache_read_input_tokens: int
+    cache_creation_input_tokens: int
     startTime: float  # Note: making this camelCase was a mistake, everything should be snake case
     endTime: float
     completionStartTime: float