Update think block extraction and format inline code execution blocks

gemini_web2api.py

@@ -507,10 +507,10 @@ def gemini_stream_generate(prompt: str, model_info: dict) -> str:
 
 def clean_gemini_text(text: str) -> str:
     """Remove internal code execution artifacts and image placeholders."""
-    # Remove code execution artifacts
+    # Convert code execution blocks to standard markdown
     text = re.sub(
-        r'```(?:python|javascript|text)\?code_(?:reference|stdout)&code_event_index=\d+\n.*?```\n?',
-        '', text, flags=re.DOTALL
+        r'```(python|javascript|text)\?code_(?:reference|stdout)&code_event_index=\d+',
+        r'```\1', text
     )
     # Remove googleusercontent placeholder URLs (image gen/retrieval/collection)
     text = re.sub(
@@ -571,7 +571,7 @@ def _scan_complete_wrb_frames(buf: str) -> list:
     return frames
 
 
-def extract_response_text(raw: str) -> str:
+def extract_response_text(raw: str, model_info: dict = None) -> str:
     """Parse StreamGenerate response to extract final text.
 
     Uses the upstream's bracket-depth frame scanner for robustness.
@@ -607,6 +607,10 @@ def extract_response_text(raw: str) -> str:
             break
             
     final_response = clean_gemini_text(text)
+    
+    if model_info and model_info.get("think") == 0:
+        final_response = f"<think>\nthink_mode: 0\nwe do not really recieve thinking block from gemini.\n</think>\n\n{final_response}"
+        
     return final_response
 
 
@@ -792,7 +796,7 @@ class GeminiHandler(BaseHTTPRequestHandler):
 
     def _call_gemini(self, prompt, model_info, tools):
         raw = gemini_stream_generate(prompt, model_info)
-        text = extract_response_text(raw)
+        text = extract_response_text(raw, model_info)
         tool_calls = None
         if tools and text:
             text, tool_calls = parse_tool_calls(text)