@@ -1113,3 +1113,6 @@ msgid "Model is blocked by outdating or temporary block, please retry later"
 msgstr ""
 "Модель заблокирована, т.к перестала обновляться или временно, попробуйте "
 "позже"
+
+msgid "The model is currently disabled. Please try again later."
+msgstr "Модель в настоящее время неактивна. Пожалуйста, повторите попытку позже."
@@ -1,6 +1,6 @@
+from django.utils.translation import gettext as _
 # накинуть перевод через gettext_lazy
 
-
 class GenerationException(Exception):
     def __str__(self):
         return 'Случилась ошибка во время генерации у этой модели, пожалуйста повторите попытку позже'
@@ -10,3 +10,8 @@ class NSFWDetectedException(Exception): ...
 
 
 class LargeResourceConsumptionException(Exception): ...
+
+
+class DeploymentDisabled(Exception):
+    def __str__(self):
+        return _('The model is currently disabled. Please try again later.')
@@ -16,6 +16,7 @@ from deepl.translator import TextResult
 from requests import Response
 
 from backend import settings
+from ml_model.exceptions import DeploymentDisabled
 from ml_model.utils import count_openrouter_tokens
 from poller.models import Proxy
 
@@ -144,12 +145,14 @@ def openrouter_run(version: str, messages: list, callback_data: dict, model_name
                     answer = f'**Рассуждение:**\n\n{reasoning}\n\n**Основная мысль:**\n\n{content}'
                 elif content:
                     answer = content
-                if data.get('choices')[0].get('error') is not None:
-                    error = json.loads(
-                        data['choices'][0]['error']['metadata'].replace("'", '"')
-                    ).get('raw', {}).get('type')
-                    if error in ('overloaded_error',):
+                if int(data.get('choices')[0].get('error', {}).get('code', 0)) == 502:
+                    error_type = re.sub(r'["\']', '', str(data['choices'][0]['error']['message']))
+                    if error_type == 'Overloaded':
+                        logger.warning(f"Model {model_name} overloaded")
                         input_tokens, output_tokens = count_openrouter_tokens(model_name, messages, content + reasoning)
+                    else:
+                        logger.error(f"Model {model_name} disabled")
+                        raise DeploymentDisabled
                 else:
                     input_tokens = data['usage']['prompt_tokens']
                     output_tokens = data['usage']['completion_tokens']
@@ -12,11 +12,13 @@ from rest_framework.response import Response
 from rest_framework.status import (
     HTTP_402_PAYMENT_REQUIRED,
     HTTP_500_INTERNAL_SERVER_ERROR,
+    HTTP_503_SERVICE_UNAVAILABLE
 )
 from rest_framework.views import APIView
 
 from messages.models import Message
 from messages.serializers import MessageSerializer
+from ml_model.exceptions import DeploymentDisabled
 from ml_model.services.base import SimpleService
 from payments.exceptions.insufficient_balance import InsufficientBalance
 from tools.chats.models import Chat
@@ -148,6 +150,13 @@ class MessagesAPIView(APIView):
             )
             try:
                 output_messages = service(chat).make(input_message)
+            except DeploymentDisabled as exc:
+                return Response(
+                    {
+                        'detail': f'{exc}'
+                    },
+                    status=HTTP_503_SERVICE_UNAVAILABLE,
+                )
             except Exception as exc:
                 input_message.is_sent = False
                 input_message.save()