From c1132d045470853a8a186d730678493f3f015dbe Mon Sep 17 00:00:00 2001
From: shotsan <shotsan@users.noreply.github.com>
Date: Tue, 8 Sep 2026 13:44:50 -0700
Subject: [PATCH] fix(convert_dict_to_response): handle empty choices list
 without raising 500 APIError (Fixes #40276)

---
 .../convert_dict_to_response.py               |  6 +-
 .../test_convert_dict_to_response.py          | 70 ++++++++++++++++++-
 2 files changed, 72 insertions(+), 4 deletions(-)

diff --git a/litellm/litellm_core_utils/llm_response_utils/convert_dict_to_response.py b/litellm/litellm_core_utils/llm_response_utils/convert_dict_to_response.py
index a6f10e1ede3..7d8072ba622 100644
--- a/litellm/litellm_core_utils/llm_response_utils/convert_dict_to_response.py
+++ b/litellm/litellm_core_utils/llm_response_utils/convert_dict_to_response.py
@@ -179,7 +179,7 @@ async def convert_to_streaming_response_async(
 
     choice_list: Final[list[StreamingChoices]] = []
 
-    if not response_object.get("choices"):
+    if "choices" not in response_object or not isinstance(response_object["choices"], Iterable):
         from litellm.exceptions import APIError
 
         raise APIError(
@@ -287,7 +287,7 @@ def convert_to_streaming_response(
     model_response_object: Final = ModelResponseStream()
     choice_list: Final[list[StreamingChoices]] = []
 
-    if not response_object.get("choices"):
+    if "choices" not in response_object or not isinstance(response_object["choices"], Iterable):
         from litellm.exceptions import APIError
 
         raise APIError(
@@ -623,7 +623,7 @@ def convert_to_model_response_object(
                 return convert_to_streaming_response(response_object=response_object)
             choice_list: Final[list[Choices]] = []
 
-            if not response_object.get("choices") or not isinstance(response_object["choices"], Iterable):
+            if "choices" not in response_object or not isinstance(response_object["choices"], Iterable):
                 from litellm.exceptions import APIError
 
                 raise APIError(
diff --git a/tests/test_litellm/litellm_core_utils/llm_response_utils/test_convert_dict_to_response.py b/tests/test_litellm/litellm_core_utils/llm_response_utils/test_convert_dict_to_response.py
index 304d732c518..5406df691aa 100644
--- a/tests/test_litellm/litellm_core_utils/llm_response_utils/test_convert_dict_to_response.py
+++ b/tests/test_litellm/litellm_core_utils/llm_response_utils/test_convert_dict_to_response.py
@@ -1,4 +1,4 @@
-
+import pytest
 
 from litellm.constants import RESPONSE_FORMAT_TOOL_NAME
 from litellm.litellm_core_utils.llm_response_utils.convert_dict_to_response import (
@@ -99,3 +99,71 @@ def test_handle_invalid_parallel_tool_calls_skips_custom_tool_calls():
     )
     result = _handle_invalid_parallel_tool_calls([custom_tool_call, function_tool_call])
     assert result == [custom_tool_call, function_tool_call]
+
+
+def test_convert_empty_choices_response() -> None:
+    from litellm.litellm_core_utils.llm_response_utils.convert_dict_to_response import (
+        convert_to_streaming_response,
+    )
+
+    resp = {
+        "id": "x",
+        "created": 1,
+        "model": "gemini-3.5-flash",
+        "object": "chat.completion",
+        "choices": [],
+        "usage": {"prompt_tokens": 10, "completion_tokens": 0, "total_tokens": 10},
+        "vertex_ai_safety_results": ["blocked"],
+    }
+    result = convert_to_model_response_object(
+        response_object=resp,
+        model_response_object=ModelResponse(),
+        response_type="completion",
+    )
+    assert result.choices == []
+    assert getattr(result, "vertex_ai_safety_results") == ["blocked"]
+
+    # Test sync streaming generator handles empty choices
+    sync_stream = list(convert_to_streaming_response(response_object=resp))
+    assert len(sync_stream) == 1
+    assert sync_stream[0].choices == []
+
+
+@pytest.mark.asyncio
+async def test_convert_empty_choices_response_async() -> None:
+    from litellm.litellm_core_utils.llm_response_utils.convert_dict_to_response import (
+        convert_to_streaming_response_async,
+    )
+
+    resp = {
+        "id": "x",
+        "created": 1,
+        "model": "gemini-3.5-flash",
+        "object": "chat.completion",
+        "choices": [],
+        "usage": {"prompt_tokens": 10, "completion_tokens": 0, "total_tokens": 10},
+    }
+    async_chunks = []
+    async for chunk in convert_to_streaming_response_async(response_object=resp):
+        async_chunks.append(chunk)
+    assert len(async_chunks) == 1
+    assert async_chunks[0].choices == []
+
+
+def test_convert_missing_choices_raises_api_error() -> None:
+    from litellm.exceptions import APIError
+
+    resp = {
+        "id": "x",
+        "created": 1,
+        "model": "gemini-3.5-flash",
+        "object": "chat.completion",
+    }
+    with pytest.raises(APIError) as exc_info:
+        convert_to_model_response_object(
+            response_object=resp,
+            model_response_object=ModelResponse(),
+            response_type="completion",
+        )
+    assert "no 'choices'" in str(exc_info.value)
+