fix: resolve type errors in OpenRouter error response handling

Author: abhishekbhakat

pydantic_ai_slim/pydantic_ai/models/_openai_compat.py

@@ -9,7 +9,7 @@
 from collections.abc import AsyncIterable, AsyncIterator, Callable, Mapping
 from dataclasses import dataclass, field, replace
 from datetime import datetime
-from typing import Any, Literal, overload
+from typing import Any, Literal, cast, overload
 
 from pydantic import ValidationError
 from typing_extensions import assert_never
@@ -27,7 +27,7 @@
     ToolCallPart,
 )
 
-from .. import UnexpectedModelBehavior, _utils, usage
+from .. import ModelHTTPError, UnexpectedModelBehavior, _utils, usage
 from .._output import OutputObjectDefinition
 from .._thinking_part import split_content_into_text_and_thinking
 from .._utils import guard_tool_call_id as _guard_tool_call_id, now_utc as _now_utc, number_to_datetime
@@ -277,6 +277,18 @@ def process_response(
     if not isinstance(response, chat.ChatCompletion):
         raise UnexpectedModelBehavior('Invalid response from OpenAI chat completions endpoint, expected JSON data')
 
+    if hasattr(response, 'error'):
+        error_attr = getattr(response, 'error', None)
+        if error_attr and isinstance(error_attr, dict):
+            error_dict = cast(dict[str, Any], error_attr)
+            error_code = error_dict.get('code')
+            status_code = error_code if isinstance(error_code, int) else 500
+            raise ModelHTTPError(
+                status_code=status_code,
+                model_name=getattr(model, 'model_name', 'unknown'),
+                body={'error': error_dict},
+            )
+
     if response.created:
         timestamp = number_to_datetime(response.created)
     else:

tests/models/test_openrouter.py

@@ -1,13 +1,13 @@
 import json
 import os
 from typing import Any, Literal, cast
-from unittest.mock import patch
+from unittest.mock import AsyncMock, MagicMock, patch
 
 import pydantic_core
 import pytest
 from inline_snapshot import snapshot
 
-from pydantic_ai import Agent
+from pydantic_ai import Agent, ModelHTTPError
 from pydantic_ai.messages import (
     ImageUrl,
     ModelMessage,
@@ -19,6 +19,8 @@
     UserPromptPart,
 )
 from pydantic_ai.models import ModelRequestParameters
+from pydantic_ai.models.openrouter import OpenRouterModel
+from pydantic_ai.providers.openrouter import OpenRouterProvider
 from pydantic_ai.settings import ModelSettings
 from pydantic_ai.tools import ToolDefinition
 
@@ -31,6 +33,7 @@
 
 with try_import() as imports_successful:
     from openai.types.chat import (
+        ChatCompletion,
         ChatCompletionChunk,
         ChatCompletionMessage,
     )
@@ -46,9 +49,6 @@
     )
     from openai.types.completion_usage import CompletionUsage
 
-    from pydantic_ai.models.openrouter import OpenRouterModel
-    from pydantic_ai.providers.openrouter import OpenRouterProvider
-
     def create_openrouter_model(model_name: str, mock_client: Any) -> OpenRouterModel:
         """Helper to create OpenRouterModel with mock client using provider pattern."""
         provider = OpenRouterProvider(openai_client=mock_client)
@@ -93,8 +93,6 @@ def chunk(choices: list[ChunkChoice]) -> ChatCompletionChunk:
 def test_openrouter_model_init():
     c = completion_message(ChatCompletionMessage(content='test', role='assistant'))
     mock_client = MockOpenAI.create_mock(c)
-    from pydantic_ai.providers.openrouter import OpenRouterProvider
-
     provider = OpenRouterProvider(openai_client=mock_client)
     model = OpenRouterModel('google/gemini-2.5-flash-lite', provider=provider)
     assert model.model_name == 'google/gemini-2.5-flash-lite'
@@ -418,9 +416,6 @@ async def test_openrouter_with_reasoning_settings(allow_model_requests: None):
 
 async def test_openrouter_model_custom_base_url(allow_model_requests: None):
     """Test OpenRouterModel with provider."""
-    # Test with provider using default base URL
-    from pydantic_ai.providers.openrouter import OpenRouterProvider
-
     provider = OpenRouterProvider(api_key='test-key')
     model = OpenRouterModel('openai/gpt-4o', provider=provider)
     assert model.model_name == 'openai/gpt-4o'
@@ -573,3 +568,35 @@ async def test_openrouter_user_prompt_mixed_content(allow_model_requests: None):
 
     assert result['role'] == 'user'
     assert result['content'] == 'Hello, here is an image:   and some more text.'
+
+
+async def test_openrouter_error_response_with_error_key(allow_model_requests: None):
+    """Test that OpenRouter error responses with 'error' key are properly handled.
+
+    Regression test for issue #2323 where OpenRouter returns HTTP 200 with an error
+    object in the body (e.g., from upstream provider failures like Chutes).
+    """
+    with patch('pydantic_ai.models.openrouter.OpenRouterProvider') as mock_provider_class:
+        mock_provider = MagicMock()
+        mock_provider.client = AsyncMock()
+        mock_provider.model_profile = MagicMock(return_value=MagicMock())
+        mock_provider_class.return_value = mock_provider
+
+        model = OpenRouterModel('deepseek/deepseek-chat-v3-0324:free')
+
+        error_response = MagicMock(spec=ChatCompletion)
+        error_response.id = None
+        error_response.choices = None
+        error_response.created = None
+        error_response.model = None
+        error_response.object = None
+        error_response.error = {'message': 'Internal Server Error', 'code': 500}
+
+        with patch('pydantic_ai.models.openrouter.completions_create', new_callable=AsyncMock) as mock_create:
+            mock_create.return_value = error_response
+
+            messages: list[ModelMessage] = [ModelRequest(parts=[UserPromptPart(content='test')])]
+            model_params = cast(ModelRequestParameters, {})
+
+            with pytest.raises(ModelHTTPError, match='status_code: 500.*Internal Server Error'):
+                await model.request(messages, None, model_params)