Add cache 5m/1h ttl write tokens from anthropic (#7779)

Yun-Kim · web-flow · commit d1b3492c7535 · 2026-03-13T21:22:04.000Z
diff --git a/packages/dd-trace/src/llmobs/constants/tags.js b/packages/dd-trace/src/llmobs/constants/tags.js
@@ -38,6 +38,8 @@ module.exports = {
   TOTAL_TOKENS_METRIC_KEY: 'total_tokens',
   CACHE_READ_INPUT_TOKENS_METRIC_KEY: 'cache_read_input_tokens',
   CACHE_WRITE_INPUT_TOKENS_METRIC_KEY: 'cache_write_input_tokens',
+  CACHE_WRITE_5M_INPUT_TOKENS_METRIC_KEY: 'ephemeral_5m_input_tokens',
+  CACHE_WRITE_1H_INPUT_TOKENS_METRIC_KEY: 'ephemeral_1h_input_tokens',
   REASONING_OUTPUT_TOKENS_METRIC_KEY: 'reasoning_output_tokens',
 
   DROPPED_IO_COLLECTION_ERROR: 'dropped_io',
diff --git a/packages/dd-trace/src/llmobs/plugins/anthropic.js b/packages/dd-trace/src/llmobs/plugins/anthropic.js
@@ -253,6 +253,15 @@ class AnthropicLLMObsPlugin extends LLMObsPlugin {
     if (cacheWriteTokens != null) metrics.cacheWriteTokens = cacheWriteTokens
     if (cacheReadTokens != null) metrics.cacheReadTokens = cacheReadTokens
 
+    const cacheCreation = usage.cache_creation
+    if (cacheCreation) {
+      metrics.cacheWrite5mTokens = cacheCreation.ephemeral_5m_input_tokens ?? 0
+      metrics.cacheWrite1hTokens = cacheCreation.ephemeral_1h_input_tokens ?? 0
+    } else if (cacheWriteTokens != null) {
+      metrics.cacheWrite5mTokens = cacheWriteTokens
+      metrics.cacheWrite1hTokens = 0
+    }
+
     this._tagger.tagMetrics(span, metrics)
   }
 
diff --git a/packages/dd-trace/src/llmobs/tagger.js b/packages/dd-trace/src/llmobs/tagger.js
@@ -22,6 +22,8 @@ const {
   ROOT_PARENT_ID,
   CACHE_READ_INPUT_TOKENS_METRIC_KEY,
   CACHE_WRITE_INPUT_TOKENS_METRIC_KEY,
+  CACHE_WRITE_5M_INPUT_TOKENS_METRIC_KEY,
+  CACHE_WRITE_1H_INPUT_TOKENS_METRIC_KEY,
   INPUT_TOKENS_METRIC_KEY,
   OUTPUT_TOKENS_METRIC_KEY,
   TOTAL_TOKENS_METRIC_KEY,
@@ -185,6 +187,12 @@ class LLMObsTagger {
         case 'cacheWriteTokens':
           processedKey = CACHE_WRITE_INPUT_TOKENS_METRIC_KEY
           break
+        case 'cacheWrite5mTokens':
+          processedKey = CACHE_WRITE_5M_INPUT_TOKENS_METRIC_KEY
+          break
+        case 'cacheWrite1hTokens':
+          processedKey = CACHE_WRITE_1H_INPUT_TOKENS_METRIC_KEY
+          break
         case 'reasoningOutputTokens':
           processedKey = REASONING_OUTPUT_TOKENS_METRIC_KEY
           break
diff --git a/packages/dd-trace/test/llmobs/plugins/anthropic/index.spec.js b/packages/dd-trace/test/llmobs/plugins/anthropic/index.spec.js
@@ -32,6 +32,8 @@ function assertLLMObsSpan (apmSpans, llmobsSpans) {
       total_tokens: MOCK_NUMBER,
       cache_write_input_tokens: MOCK_NUMBER,
       cache_read_input_tokens: MOCK_NUMBER,
+      ephemeral_5m_input_tokens: MOCK_NUMBER,
+      ephemeral_1h_input_tokens: MOCK_NUMBER,
     },
     tags: { ml_app: 'test', integration: 'anthropic' },
   })
