From b30e3e214d655134ec525005396ddaec67928d53 Mon Sep 17 00:00:00 2001
From: Florian Reifschneider <florian@ag.dev>
Date: Tue, 14 Apr 2026 13:06:51 -0700
Subject: [PATCH 1/2] fix(langfuse): avoid double-counting cache and reasoning
 tokens

Langfuse sums all keys containing "input" for Input usage and "output"
for Output usage. Since inputTokens/outputTokens already included
cache/reasoning tokens, they were double-counted. Now input/output
report only their non-overlapping portion, and total is omitted so
Langfuse derives it correctly by summing all keys.
---
 packages/langfuse/src/index.test.ts |  6 ------
 packages/langfuse/src/index.ts      | 13 +++++++------
 2 files changed, 7 insertions(+), 12 deletions(-)

diff --git a/packages/langfuse/src/index.test.ts b/packages/langfuse/src/index.test.ts
index 6c11ebd..2c6c178 100644
--- a/packages/langfuse/src/index.test.ts
+++ b/packages/langfuse/src/index.test.ts
@@ -238,7 +238,6 @@ describe('@core-ai/langfuse', () => {
             usageDetails: {
                 input: 5,
                 output: 3,
-                total: 8,
                 cache_read_input: 0,
                 cache_creation_input: 0,
             },
@@ -273,7 +272,6 @@ describe('@core-ai/langfuse', () => {
             usageDetails: {
                 input: 5,
                 output: 3,
-                total: 8,
                 cache_read_input: 0,
                 cache_creation_input: 0,
             },
@@ -314,7 +312,6 @@ describe('@core-ai/langfuse', () => {
             usageDetails: {
                 input: 5,
                 output: 3,
-                total: 8,
                 cache_read_input: 0,
                 cache_creation_input: 0,
             },
@@ -364,7 +361,6 @@ describe('@core-ai/langfuse', () => {
             usageDetails: {
                 input: 5,
                 output: 3,
-                total: 8,
                 cache_read_input: 0,
                 cache_creation_input: 0,
             },
@@ -491,7 +487,6 @@ describe('@core-ai/langfuse', () => {
             usageDetails: {
                 input: 5,
                 output: 3,
-                total: 8,
                 cache_read_input: 0,
                 cache_creation_input: 0,
             },
@@ -539,7 +534,6 @@ describe('@core-ai/langfuse', () => {
         expect(observation.update).toHaveBeenNthCalledWith(2, {
             usageDetails: {
                 input: 3,
-                total: 3,
             },
         });
     });
diff --git a/packages/langfuse/src/index.ts b/packages/langfuse/src/index.ts
index 58063ee..3c6aee4 100644
--- a/packages/langfuse/src/index.ts
+++ b/packages/langfuse/src/index.ts
@@ -107,12 +107,14 @@ function createImageModelParameters(
 }
 
 function createChatUsageDetails(usage: ChatUsage): Record<string, number> {
+    const { cacheReadTokens, cacheWriteTokens } = usage.inputTokenDetails;
+    const reasoningTokens = usage.outputTokenDetails.reasoningTokens ?? 0;
+
     return {
-        input: usage.inputTokens,
-        output: usage.outputTokens,
-        total: usage.inputTokens + usage.outputTokens,
-        cache_read_input: usage.inputTokenDetails.cacheReadTokens,
-        cache_creation_input: usage.inputTokenDetails.cacheWriteTokens,
+        input: usage.inputTokens - cacheReadTokens - cacheWriteTokens,
+        output: usage.outputTokens - reasoningTokens,
+        cache_read_input: cacheReadTokens,
+        cache_creation_input: cacheWriteTokens,
         ...(usage.outputTokenDetails.reasoningTokens !== undefined
             ? { reasoning_output: usage.outputTokenDetails.reasoningTokens }
             : {}),
@@ -128,7 +130,6 @@ function createEmbeddingUsageDetails(
 
     return {
         input: usage.inputTokens,
-        total: usage.inputTokens,
     };
 }
 

From 90cf021f24970603439c30045e2668c0c4e18242 Mon Sep 17 00:00:00 2001
From: Florian Reifschneider <florian@ag.dev>
Date: Tue, 14 Apr 2026 13:08:32 -0700
Subject: [PATCH 2/2] add changeset

---
 .changeset/fix-langfuse-usage-double-counting.md | 5 +++++
 1 file changed, 5 insertions(+)
 create mode 100644 .changeset/fix-langfuse-usage-double-counting.md

diff --git a/.changeset/fix-langfuse-usage-double-counting.md b/.changeset/fix-langfuse-usage-double-counting.md
new file mode 100644
index 0000000..28c3944
--- /dev/null
+++ b/.changeset/fix-langfuse-usage-double-counting.md
@@ -0,0 +1,5 @@
+---
+'@core-ai/langfuse': patch
+---
+
+Fix double-counting of cache and reasoning tokens in Langfuse usage breakdown. The `input` and `output` keys now report only non-overlapping token counts so Langfuse's aggregation sums correctly.