fix(interactions): raise on a non-2xx Gemini interaction fetch

AsyncHTTPHandler.get never raises for status and the Gemini GET transform
only raised when the body was not JSON, so a 500 or 404 carrying Gemini's
JSON error body parsed as an interaction with no status. A delete on a
replica other than the creator then claimed the settlement as released and
forwarded the delete instead of failing closed, and the bill was lost. The
transform now raises GeminiError with the vendor's status, as the delete
transform already does; the in-process poll already retries a fetch that
raises

litellm/llms/gemini/interactions/transformation.py

@@ -313,6 +313,12 @@ class GoogleAIStudioInteractionsConfig(BaseInteractionsAPIConfig):
         raw_response: httpx.Response,
         logging_obj: LiteLLMLoggingObj,
     ) -> InteractionsAPIResponse:
+        if not 200 <= raw_response.status_code < 300:
+            raise GeminiError(
+                message=raw_response.text,
+                status_code=raw_response.status_code,
+                headers=dict(raw_response.headers),
+            )
         try:
             raw_json: Final = _interaction_body(raw_response)
         except Exception:

tests/unit/interactions/test_gemini_interactions_transformation.py

@@ -10,12 +10,13 @@ Covers:
 
 from unittest.mock import MagicMock, patch
 
+import httpx
 import pytest
 
-
 from litellm.interactions.litellm_responses_transformation.streaming_iterator import (
     LiteLLMResponsesInteractionsStreamingIterator,
 )
+from litellm.llms.gemini.common_utils import GeminiError
 from litellm.llms.gemini.interactions.transformation import (
     GoogleAIStudioInteractionsConfig,
 )
@@ -464,6 +465,32 @@ class TestInteractionOperationUrls:
                 )
 
 
+class TestGetInteractionResponse:
+    @pytest.mark.parametrize("status_code", [404, 500])
+    def test_non_2xx_raises_even_when_the_error_body_is_json(
+        self, config: GoogleAIStudioInteractionsConfig, status_code: int
+    ) -> None:
+        raw_response = httpx.Response(
+            status_code,
+            json={"error": {"code": status_code, "message": "boom", "status": "INTERNAL"}},
+            request=httpx.Request("GET", "https://generativelanguage.googleapis.com/v1beta/interactions/x"),
+        )
+        with pytest.raises(GeminiError) as raised:
+            config.transform_get_interaction_response(raw_response=raw_response, logging_obj=MagicMock())
+        assert raised.value.status_code == status_code
+        assert "boom" in str(raised.value)
+
+    def test_2xx_parses_the_interaction(self, config: GoogleAIStudioInteractionsConfig) -> None:
+        raw_response = httpx.Response(
+            200,
+            json={"id": "interaction-1", "object": "interaction", "status": "completed", "steps": []},
+            request=httpx.Request("GET", "https://generativelanguage.googleapis.com/v1beta/interactions/x"),
+        )
+        response = config.transform_get_interaction_response(raw_response=raw_response, logging_obj=MagicMock())
+        assert response.id == "interaction-1"
+        assert response.status == "completed"
+
+
 class TestTransformRequestSchemaCoalescing:
     """Test new-schema request coalescing (Api-Revision: 2026-05-20)."""