Wire through token cache statistics

ammario · ammario · commit 00dc34fcf61a · 2025-08-11T21:46:31.000-05:00
diff --git a/src/anthropic-api-types.ts b/src/anthropic-api-types.ts
@@ -118,6 +118,8 @@ export type AnthropicToolChoice =
 export type AnthropicStreamUsage = {
   input_tokens: number;
   output_tokens: number;
+  cache_creation_input_tokens?: number;
+  cache_read_input_tokens?: number;
 };
 
 export type AnthropicStreamChunk =
diff --git a/src/anthropic-proxy.ts b/src/anthropic-proxy.ts
@@ -325,6 +325,13 @@ export const createAnthropicProxy = ({
                   usage: {
                     input_tokens: usage.inputTokens,
                     output_tokens: usage.outputTokens,
+                    // OpenAI provides cached tokens via cachedInputTokens or in experimental_providerMetadata
+                    // Map to Anthropic's cache_read_input_tokens
+                    cache_creation_input_tokens: 0, // OpenAI doesn't report cache creation separately
+                    cache_read_input_tokens: usage.cachedInputTokens ?? 
+                                            (typeof (response as any).experimental_providerMetadata?.openai?.cached_tokens === 'number'
+                                              ? (response as any).experimental_providerMetadata.openai.cached_tokens
+                                              : 0),
                   },
                 })
               );
diff --git a/src/convert-to-anthropic-stream.ts b/src/convert-to-anthropic-stream.ts
@@ -27,7 +27,12 @@ export function convertToAnthropicStream(
               model: "claude-4-sonnet-20250514",
               stop_reason: null,
               stop_sequence: null,
-              usage: { input_tokens: 0, output_tokens: 0 },
+              usage: {
+                input_tokens: 0,
+                output_tokens: 0,
+                cache_creation_input_tokens: 0,
+                cache_read_input_tokens: 0,
+              },
             },
           });
           break;
@@ -42,6 +47,14 @@ export function convertToAnthropicStream(
             usage: {
               input_tokens: chunk.usage.inputTokens ?? 0,
               output_tokens: chunk.usage.outputTokens ?? 0,
+              // OpenAI provides cached tokens via cachedInputTokens or in providerMetadata
+              cache_creation_input_tokens: 0, // OpenAI doesn't report cache creation separately
+              cache_read_input_tokens:
+                chunk.usage.cachedInputTokens ??
+                (typeof chunk.providerMetadata?.openai?.cached_tokens ===
+                "number"
+                  ? chunk.providerMetadata.openai.cached_tokens
+                  : 0),
             },
           });
           break;
