@@ -56,6 +56,11 @@ class Gemini(SimpleService):
             'input_imgs': Decimal('0'),
             'highest_prices': {'input': Decimal('1200'), 'output': Decimal('5400')},
         },
+        'gemini-3-flash-preview': {
+            'input': Decimal('150'),
+            'output': Decimal('900'),
+            'input_imgs': Decimal('0'),
+        },
     }
 
     TOOLS_TOKEN_COSTS = {'text-embedding-3-large': {'output': Decimal('0.000065')}}
@@ -77,7 +82,6 @@ class Gemini(SimpleService):
         if image:
             price += price_map['input_imgs'] / 1_000
         if embedding_tokens > 0:
-            print(embedding_tokens)
             price += self.TOOLS_TOKEN_COSTS['text-embedding-3-large']['output'] * embedding_tokens
         return price.quantize(Decimal('0.1'), rounding='ROUND_UP')
 
@@ -102,6 +106,38 @@ class Gemini(SimpleService):
         }
         messages = self.get_chat_history()
         messages.append({'role': 'user', 'content': input_message.content})
+        if version == 'google/gemini-3-flash-preview':
+            messages.insert(
+                0,
+                {
+                    'role': 'system',
+                    'content': 'You are operating in Deep Analytical Reasoning Mode. Your goal is to approximate '
+                    'research-grade reasoning depth similar to advanced long-thinking models while '
+                    'maintaining accuracy, structure, and verification. CORE DIRECTIVES: 1. Decompose '
+                    'every complex problem before answering — identify knowns, unknowns, constraints, '
+                    'and assumptions; break tasks into sub-problems. 2. Use Multi-Hypothesis Reasoning — '
+                    'generate multiple solution paths and explore 2–3 strategies when complexity is high. '
+                    '3. Apply Step-by-Step Logical Derivation — show intermediate reasoning and justify '
+                    'each transition logically or mathematically. 4. Perform Cross-Validation — re-check '
+                    'conclusions using alternative logic, formulas, or perspectives and detect '
+                    'contradictions. 5. Run an Error Detection Loop — reassess derived answers, '
+                    'question possible mistakes, and revise if needed. 6. Evidence-Bound Reasoning Only — '
+                    'base conclusions strictly on provided data or established knowledge; state uncertainty '
+                    'explicitly. DEPTH SCALING: Automatically increase reasoning depth for mathematics, '
+                    'algorithms, system design, scientific analysis, financial modeling, legal reasoning, '
+                    'and architecture planning. STRUCTURED OUTPUT FORMAT for complex tasks: Problem '
+                    'Decomposition → Variables & Constraints → Hypothesis Generation → Step-by-Step '
+                    'Reasoning → Cross-Validation → Final Answer → Confidence Level with justification. '
+                    'ANTI-SHALLOW RULES: Do not skip reasoning steps, avoid surface-level summaries, '
+                    'avoid intuition-only answers, prefer rigor over brevity. SELF-REFLECTION DIRECTIVE: '
+                    'Review the reasoning chain before finalizing, identify gaps, and strengthen weak logic. '
+                    'Priority: analytical depth, internal consistency, and correctness over speed.',
+                },
+            )
+            callback_data.update({
+                "reasoning": {"effort": "high"},
+                "temperature": 0.2
+            })
         file = input_message.file
         image = None
         embedding_tokens = 0
@@ -13,15 +13,15 @@ from tools.public_api.models import APIStore
 class Qwen_3_Max_Thinking(SimpleService):
     TOKENS_COST = {
         'input': {'default': Decimal('360'), 'high': Decimal('900')},
-        'output': {'default': Decimal('1800'), 'high': Decimal('4500')}
+        'output': {'default': Decimal('1800'), 'high': Decimal('4500')},
     }
 
     def calculate_price(self, input_tokens: int, output_tokens: int) -> Decimal:
-        price = (
-            input_tokens * (self.TOKENS_COST['input']['default' if input_tokens <= 128_000 else 'high'] / 1_000_000)
-            + output_tokens * (self.TOKENS_COST['output']['default' if input_tokens <= 128_000 else 'high'] / 1_000_000)
-            + Decimal('6')
-        )
+        price = input_tokens * (
+            self.TOKENS_COST['input']['default' if input_tokens <= 32_000 else 'high'] / 1_000_000
+        ) + output_tokens * (
+            self.TOKENS_COST['output']['default' if input_tokens <= 32_000 else 'high'] / 1_000_000
+        ) + Decimal('2')
         return price.quantize(Decimal('0.1'), rounding='ROUND_UP')
 
     def save_results(self, content: str, time: timedelta, save: bool = True) -> list[Message]:
@@ -41,7 +41,7 @@ class Qwen_3_Max_Thinking(SimpleService):
         callback_data = {'provider': {'order': ['alibaba']}, **input_message.info}
         messages = self.get_chat_history()
         messages.append({'role': 'user', 'content': input_message.content})
-        result = openrouter_run('qwen/qwen3-max:online', messages, callback_data, 'Qwen')
+        result = openrouter_run('qwen/qwen3-max-thinking:online', messages, callback_data, 'Qwen')
         process_time = timedelta(seconds=(time.time() - start_time))
         self.handle_invoice(
             input_message.content_object.model,
@@ -51,13 +51,15 @@ class Qwen_3_Max_Thinking(SimpleService):
         msgs = self.save_results(result[0], process_time)
         return msgs
 
-    def get_chat_history(self, message_limit: int = 10, max_character_limit: int = 1500) -> list[dict[str, str | list]]:
+    def get_chat_history(
+        self, message_limit: int = 10, max_character_limit: int = 1500
+    ) -> list[dict[str, str | list]]:
         if isinstance(self.store, Chat):
             air_messages = list(
                 reversed(
                     Message.objects.filter(
                         chats_chats_messages=self.store, is_deleted=False, is_sent=True
-                    ).order_by('-created_at')[1:message_limit+1]
+                    ).order_by('-created_at')[1 : message_limit + 1]
                 )
             )
         elif isinstance(self.store, APIStore):