From 9fbd6e619dec4a5f5635eb619a01d57c51827141 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Thi=C3=AAn=20To=C3=A1n?= <toanalien@gmail.com>
Date: Fri, 20 Feb 2026 14:24:21 +0700
Subject: [PATCH] fix: correct token extraction for Claude non-streaming
 responses (#131)

- Add response logging for non-streaming requests (5_res_provider.json, 7_res_client.json)
- Fix extractUsageFromResponse() to check Claude format before OpenAI format
- Prevents format misidentification that caused tokens to show as 0
- Claude uses input_tokens/output_tokens vs OpenAI's prompt_tokens/completion_tokens

Fixes dashboard Details tab showing 0 tokens for Claude requests
---
 open-sse/handlers/chatCore.js | 35 +++++++++++++++++++++++------------
 1 file changed, 23 insertions(+), 12 deletions(-)

diff --git a/open-sse/handlers/chatCore.js b/open-sse/handlers/chatCore.js
index cd774630..dee1c18b 100644
--- a/open-sse/handlers/chatCore.js
+++ b/open-sse/handlers/chatCore.js
@@ -194,18 +194,8 @@ function translateNonStreamingResponse(responseBody, targetFormat, sourceFormat)
 function extractUsageFromResponse(responseBody, provider) {
   if (!responseBody || typeof responseBody !== 'object') return null;
 
-  // OpenAI format
-  if (responseBody.usage && typeof responseBody.usage === 'object') {
-    return {
-      prompt_tokens: responseBody.usage.prompt_tokens || 0,
-      completion_tokens: responseBody.usage.completion_tokens || 0,
-      cached_tokens: responseBody.usage.prompt_tokens_details?.cached_tokens,
-      reasoning_tokens: responseBody.usage.completion_tokens_details?.reasoning_tokens
-    };
-  }
-
-  // Claude format
-  if (responseBody.usage && typeof responseBody.usage === 'object' && (responseBody.usage.input_tokens !== undefined || responseBody.usage.output_tokens !== undefined)) {
+  // Claude format - check first to avoid conflict with OpenAI check
+  if (responseBody.usage && typeof responseBody.usage === 'object' && responseBody.usage.input_tokens !== undefined) {
     return {
       prompt_tokens: responseBody.usage.input_tokens || 0,
       completion_tokens: responseBody.usage.output_tokens || 0,
@@ -214,6 +204,16 @@ function extractUsageFromResponse(responseBody, provider) {
     };
   }
 
+  // OpenAI format
+  if (responseBody.usage && typeof responseBody.usage === 'object' && responseBody.usage.prompt_tokens !== undefined) {
+    return {
+      prompt_tokens: responseBody.usage.prompt_tokens || 0,
+      completion_tokens: responseBody.usage.completion_tokens || 0,
+      cached_tokens: responseBody.usage.prompt_tokens_details?.cached_tokens,
+      reasoning_tokens: responseBody.usage.completion_tokens_details?.reasoning_tokens
+    };
+  }
+
   // Gemini format
   if (responseBody.usageMetadata && typeof responseBody.usageMetadata === 'object') {
     return {
@@ -723,6 +723,14 @@ export async function handleChatCore({ body, modelInfo, credentials, log, onCred
       }
     }
 
+    // Log provider response (raw response from provider)
+    reqLogger.logProviderResponse(
+      providerResponse.status,
+      providerResponse.statusText,
+      providerResponse.headers,
+      responseBody
+    );
+
     // Notify success - caller can clear error status if needed
     if (onRequestSuccess) {
       await onRequestSuccess();
@@ -758,6 +766,9 @@ export async function handleChatCore({ body, modelInfo, credentials, log, onCred
       translatedResponse.usage = filterUsageForFormat(buffered, sourceFormat);
     }
 
+    // Log converted response (final response to client)
+    reqLogger.logConvertedResponse(translatedResponse);
+
     const totalLatency = Date.now() - requestStartTime;
     const requestDetail = {
       provider: provider || "unknown",