fix: thinking mode was hardcoded to always-on for all models

gemini_web2api.py

@@ -74,29 +74,30 @@ MODEL_HEADER_KEY = "x-goog-ext-525001261-jspb"
 
 GEMINI_MODELS = {
     # Internal model name → routing info
+    # think: 0 = thinking enabled (for thinking models), 4 = thinking disabled (for standard models)
     "gemini-3-pro": {
-        "id": "9d8ca3786ebdfbea", "capacity": 1,
+        "id": "9d8ca3786ebdfbea", "capacity": 1, "think": 4,
         "desc": "Pro model (free tier)",
     },
     "gemini-3-flash": {
-        "id": "fbb127bbb056c959", "capacity": 1,
+        "id": "fbb127bbb056c959", "capacity": 1, "think": 4,
         "desc": "Fast general-purpose model",
     },
     "gemini-3-flash-thinking": {
-        "id": "5bf011840784117a", "capacity": 1,
+        "id": "5bf011840784117a", "capacity": 1, "think": 0,
         "desc": "Deep thinking mode",
     },
     # Pro-only (paid tier) models
     "gemini-3-pro-plus": {
-        "id": "e6fa609c3fa255c0", "capacity": 4,
+        "id": "e6fa609c3fa255c0", "capacity": 4, "think": 4,
         "desc": "Pro+ model (requires subscription)",
     },
     "gemini-3-flash-plus": {
-        "id": "56fdd199312815e2", "capacity": 4,
+        "id": "56fdd199312815e2", "capacity": 4, "think": 4,
         "desc": "Flash+ model (requires subscription)",
     },
     "gemini-3-flash-thinking-plus": {
-        "id": "e051ce1aa80aa576", "capacity": 4,
+        "id": "e051ce1aa80aa576", "capacity": 4, "think": 0,
         "desc": "Thinking+ model (requires subscription)",
     },
 }
@@ -391,8 +392,9 @@ def gemini_stream_generate(prompt: str, model_info: dict) -> str:
     inner[10] = 1
     inner[11] = 0
     # Think mode: 0 = thinking enabled, 4 = thinking disabled
-    # For thinking models, we enable thinking; for others, disable
-    inner[17] = [[0]]  # Default: thinking enabled
+    # Each model carries its own think value
+    think_mode = model_info.get("think", 4)
+    inner[17] = [[think_mode]]
     inner[18] = 0
     inner[27] = 1
     inner[30] = [4]