@@ -60,7 +60,7 @@ from ml_model.services.pulid import Pulid
 from ml_model.services.qwen import Qwen
 from ml_model.services.qwen_235B import Qwen_235B
 from ml_model.services.qwen_3_6 import Qwen_3_6
-from ml_model.services.qwen_3_7 import Qwen_3_7
+from ml_model.services.qwen_3_8 import Qwen_3_8
 from ml_model.services.qwen_3_max_thinking import Qwen_3_Max_Thinking
 from ml_model.services.raifgpt import Raifgpt
 from ml_model.services.ray import Ray
@@ -32,6 +32,7 @@ class Grok(SerperMixin, StreamSimpleService):
     TOKENS_COST = {
         'grok-4.3': {'input': Decimal('875'), 'output': Decimal('1750'), 'coefficient': Decimal('5')},
         'grok-4.5': {'input': Decimal('1000'), 'output': Decimal('3000'), 'coefficient': Decimal('5')},
+        'grok-4.6': {'input': Decimal('600'), 'output': Decimal('1800'), 'coefficient': Decimal('5')},
     }
 
     TOOLS_TOKEN_COSTS = {'text-embedding-3-small': {'output': Decimal('0.00001')}}
@@ -194,10 +195,7 @@ class Grok(SerperMixin, StreamSimpleService):
                 if (chunks_length := sum(len(chunk) for chunk in chunks)) > 20_000:
                     predict_embedding_tokens = len(chunks) * 2020
                     predicted_input_price += (
-                        (
-                            Decimal('210')
-                            + Decimal(chunks_length) / Decimal(len(chunks)) * Decimal('10')
-                        )
+                        (Decimal('210') + Decimal(chunks_length) / Decimal(len(chunks)) * Decimal('10'))
                         / Decimal('2.0')
                         * self.TOKENS_COST[version]['input']
                         / Decimal('1_000_000')
@@ -251,21 +249,14 @@ class Grok(SerperMixin, StreamSimpleService):
         if image:
             predicted_image_tokens = min((image_width * image_height + 999) // 1000, 2500)
             predicted_input_price += (
-                Decimal(predicted_image_tokens)
-                * self.TOKENS_COST[version]['input']
-                / Decimal('1_000_000')
+                Decimal(predicted_image_tokens) * self.TOKENS_COST[version]['input'] / Decimal('1_000_000')
             )
-        estimated_input_tokens = (
-            Decimal(
-                sum(
-                    len(message['content']) if isinstance(message['content'], str) else 0
-                    for message in messages
-                )
-                + (len(input_message.content) if image else 0)
+        estimated_input_tokens = Decimal(
+            sum(
+                len(message['content']) if isinstance(message['content'], str) else 0 for message in messages
             )
-            / Decimal('2.0')
-            + (150 if is_free_plan else 250)
-        )
+            + (len(input_message.content) if image else 0)
+        ) / Decimal('2.0') + (150 if is_free_plan else 250)
         predicted_input_price += (
             estimated_input_tokens * self.TOKENS_COST[version]['input'] / Decimal('1_000_000')
             + predict_embedding_tokens * self.TOOLS_TOKEN_COSTS['text-embedding-3-small']['output']
@@ -22,15 +22,16 @@ from tools.copywrite.models import Copywrite
 from tools.public_api.models import APIStore
 
 
-class Qwen_3_7(StreamSimpleService):
+class Qwen_3_8(StreamSimpleService):
     COEFFICIENT = Decimal('300.0')
 
     TOKENS_COST = {
+        'qwen3.8-max': {'input': Decimal('600'), 'output': Decimal('1800')},  # $2 / $6
         'qwen3.7-max': {'input': Decimal('442.5'), 'output': Decimal('1327.5')},  # $1.475 / $4.425
         'qwen3.7-plus': {'input': Decimal('96'), 'output': Decimal('384')},  # $0.32 / $1.28
     }
 
-    MAX_OUTPUT_TOKENS = 30_000
+    MAX_OUTPUT_TOKENS = 131_072 // 2
 
     TOOLS_TOKEN_COSTS = {'text-embedding-3-small': {'output': Decimal('0.00001')}}
 
@@ -80,7 +81,7 @@ class Qwen_3_7(StreamSimpleService):
         version_slug, model_slug, callback_data, messages, embedding_tokens = self._prepare_data(
             input_message
         )
-        result = openrouter_run(model_slug, messages, callback_data, 'Qwen 3.7')
+        result = openrouter_run(model_slug, messages, callback_data, 'Qwen 3.8')
         process_time = timedelta(seconds=(time.time() - start_time))
         self.handle_invoice(
             input_message.content_object.model,
@@ -245,7 +245,7 @@ def openrouter_run(version: str, messages: list, callback_data: dict, model_name
                         logger.error(f'Model {model_name} disabled')
                         raise DeploymentDisabled
                 else:
-                    if re.match(r'^qwen/qwen3\.7-.*$', data['model']):
+                    if re.match(r'^qwen/qwen3\.[78]-.*$', data['model']):
                         input_tokens = data['usage']['cost']
                         output_tokens = 0
                     else: