diff --git a/backend/open_webui/utils/middleware.py b/backend/open_webui/utils/middleware.py
index 7941d80790..3dde15af21 100644
--- a/backend/open_webui/utils/middleware.py
+++ b/backend/open_webui/utils/middleware.py
@@ -3623,7 +3623,31 @@ async def non_streaming_chat_response_handler(response, ctx):
                     # otherwise generate from response content
                     response_output = response_data.get('output')
                     if not response_output:
-                        response_output = [
+                        choice_message = choices[0].get('message', {})
+                        reasoning_content = choice_message.get('reasoning_content') or choice_message.get('reasoning')
+                        reasoning_details = choice_message.get('reasoning_details')
+                        response_output = []
+                        if reasoning_content or reasoning_details:
+                            reasoning_item = {
+                                'type': 'reasoning',
+                                'id': output_id('r'),
+                                'status': 'completed',
+                                'start_tag': '<think>',
+                                'end_tag': '</think>',
+                                'attributes': {'type': 'reasoning_content'},
+                                'content': (
+                                    [{'type': 'output_text', 'text': reasoning_content}]
+                                    if reasoning_content
+                                    else []
+                                ),
+                                'summary': None,
+                            }
+                            if reasoning_details:
+                                reasoning_item['reasoning_details'] = (
+                                    reasoning_details if isinstance(reasoning_details, list) else [reasoning_details]
+                                )
+                            response_output.append(reasoning_item)
+                        response_output.append(
                             {
                                 'type': 'message',
                                 'id': output_id('msg'),
@@ -3631,7 +3655,7 @@ async def non_streaming_chat_response_handler(response, ctx):
                                 'role': 'assistant',
                                 'content': [{'type': 'output_text', 'text': content}],
                             }
-                        ]
+                        )
 
                     await event_emitter(
                         {
@@ -4398,7 +4422,8 @@ async def streaming_chat_response_handler(response, ctx):
                                         or delta.get('reasoning')
                                         or delta.get('thinking')
                                     )
-                                    if reasoning_content:
+                                    reasoning_details = delta.get('reasoning_details')
+                                    if reasoning_content or reasoning_details:
                                         if not output or output[-1].get('type') != 'reasoning':
                                             reasoning_item = {
                                                 'type': 'reasoning',
@@ -4415,20 +4440,28 @@ async def streaming_chat_response_handler(response, ctx):
                                         else:
                                             reasoning_item = output[-1]
 
-                                        # Append to reasoning content
-                                        parts = reasoning_item.get('content', [])
-                                        if parts and parts[-1].get('type') == 'output_text':
-                                            parts[-1]['text'] += reasoning_content
-                                        else:
-                                            reasoning_item['content'] = [
-                                                {
-                                                    'type': 'output_text',
-                                                    'text': reasoning_content,
-                                                }
-                                            ]
+                                        if reasoning_content:
+                                            # Append to reasoning content
+                                            parts = reasoning_item.get('content', [])
+                                            if parts and parts[-1].get('type') == 'output_text':
+                                                parts[-1]['text'] += reasoning_content
+                                            else:
+                                                reasoning_item['content'] = [
+                                                    {
+                                                        'type': 'output_text',
+                                                        'text': reasoning_content,
+                                                    }
+                                                ]
 
-                                        data = {'content': serialize_output(full_output())}
-                                        delta_type = 'content'
+                                            data = {'content': serialize_output(full_output())}
+                                            delta_type = 'content'
+
+                                        if reasoning_details:
+                                            reasoning_item.setdefault('reasoning_details', []).extend(
+                                                reasoning_details
+                                                if isinstance(reasoning_details, list)
+                                                else [reasoning_details]
+                                            )
 
                                     if value:
                                         if (
diff --git a/backend/open_webui/utils/misc.py b/backend/open_webui/utils/misc.py
index c5c6528d10..af7c7e0ae9 100644
--- a/backend/open_webui/utils/misc.py
+++ b/backend/open_webui/utils/misc.py
@@ -250,10 +250,16 @@ def convert_output_to_messages(
     pending_tool_calls = []
     pending_content = []
     pending_reasoning = []  # Only populated when reasoning_format == 'reasoning_content'
+    pending_reasoning_details = []
 
     def flush_pending():
-        nonlocal pending_content, pending_tool_calls, pending_reasoning
-        if not pending_content and not pending_tool_calls and not pending_reasoning:
+        nonlocal pending_content, pending_tool_calls, pending_reasoning, pending_reasoning_details
+        if (
+            not pending_content
+            and not pending_tool_calls
+            and not pending_reasoning
+            and not pending_reasoning_details
+        ):
             return
 
         message = {
@@ -265,10 +271,14 @@ def convert_output_to_messages(
         if pending_reasoning:
             message['reasoning_content'] = '\n'.join(pending_reasoning)
 
+        if pending_reasoning_details:
+            message['reasoning_details'] = pending_reasoning_details
+
         messages.append(message)
         pending_content = []
         pending_tool_calls = []
         pending_reasoning = []
+        pending_reasoning_details = []
 
     for item in output:
         item_type = item.get('type', '')
@@ -339,7 +349,8 @@ def convert_output_to_messages(
                 )
 
         elif item_type == 'reasoning':
-            if not reasoning_format:
+            reasoning_details = item.get('reasoning_details') if raw else None
+            if not reasoning_format and not reasoning_details:
                 continue
 
             reasoning_text = ''
@@ -360,6 +371,11 @@ def convert_output_to_messages(
                     # llama.cpp: collect for reasoning_content field
                     pending_reasoning.append(reasoning_text)
 
+            if reasoning_details:
+                pending_reasoning_details.extend(
+                    reasoning_details if isinstance(reasoning_details, list) else [reasoning_details]
+                )
+
         elif item_type == 'open_webui:code_interpreter':
             # Always include code interpreter content so the LLM knows
             # the code was already executed and doesn't retry.