diff --git a/backend/open_webui/utils/context_compaction.py b/backend/open_webui/utils/context_compaction.py
index 19d44f417d..d3955148cb 100644
--- a/backend/open_webui/utils/context_compaction.py
+++ b/backend/open_webui/utils/context_compaction.py
@@ -321,12 +321,10 @@ def _exceeds_token_threshold(messages: list[dict], system_prompt: str, summary:
     if threshold <= 0:
         return False
 
-    # Expanded tool histories include fresh results and rebuilt prompts absent from prior usage.
-    if not any(message.get('role') == 'tool' for message in messages):
-        for idx in range(len(messages) - 1, -1, -1):
-            usage = messages[idx].get('usage') or (messages[idx].get('info') or {}).get('usage')
-            if isinstance(usage, dict) and (tokens := _usage_token_count(usage)):
-                return tokens + _estimate_messages_tokens(messages[idx + 1 :]) > threshold
+    for idx in range(len(messages) - 1, -1, -1):
+        usage = messages[idx].get('usage') or (messages[idx].get('info') or {}).get('usage')
+        if isinstance(usage, dict) and (tokens := _usage_token_count(usage)):
+            return tokens + _estimate_messages_tokens(messages[idx + 1 :]) > threshold
 
     estimated = _estimate_tokens(system_prompt) + _estimate_tokens(summary or '') + _estimate_messages_tokens(messages)
     return estimated > threshold
diff --git a/backend/open_webui/utils/middleware.py b/backend/open_webui/utils/middleware.py
index 57ac2e7517..1adfba67fd 100644
--- a/backend/open_webui/utils/middleware.py
+++ b/backend/open_webui/utils/middleware.py
@@ -5917,6 +5917,7 @@ async def streaming_chat_response_handler(response, ctx):
                 )
                 tool_call_sources = []  # Track citation sources from tool results
                 all_tool_call_sources = []  # Accumulated sources across all iterations
+                attached_usage = None
                 user_message = get_last_user_message(form_data['messages'])
                 messages = list(form_data['messages'])
 
@@ -6280,6 +6281,11 @@ async def streaming_chat_response_handler(response, ctx):
                                             image_urls.append(part.get('image_url', ''))
                                     message['content'] = ''.join(text_parts)
 
+                            # A count reported for the last request covers everything before these results.
+                            if usage is not attached_usage and tool_messages:
+                                tool_messages[0]['usage'] = usage
+                                attached_usage = usage
+
                             # Refresh citation context without restoring compacted history.
                             user_prompt = get_last_user_message_item(form_data['messages'])
                             for message in messages: