fix(passthrough): keep upstream error body readable for streaming error status mapping

litellm/model_prices_and_context_window_backup.json

@@ -24342,7 +24342,7 @@
         "supports_response_schema": true,
         "supports_vision": true
     },
-    "gigachat/GigaChat-2-Lite": {
+    "gigachat/GigaChat-2": {
         "input_cost_per_token": 0.0,
         "litellm_provider": "gigachat",
         "max_input_tokens": 128000,
@@ -24404,6 +24404,15 @@
         "output_cost_per_token": 0.0,
         "output_vector_size": 2560
     },
+    "gigachat/GigaEmbeddings-3B-2025-09": {
+        "input_cost_per_token": 0.0,
+        "litellm_provider": "gigachat",
+        "max_input_tokens": 4096,
+        "max_tokens": 4096,
+        "mode": "embedding",
+        "output_cost_per_token": 0.0,
+        "output_vector_size": 2048
+    },
     "gmi/anthropic/claude-opus-4.5": {
         "input_cost_per_token": 5e-06,
         "litellm_provider": "gmi",

litellm/passthrough/main.py

@@ -86,6 +86,10 @@ class AsyncPassthroughStreamingResponse(AsyncGenerator[Any, Any]):
                     self._response.raise_for_status()
                     self._iterator = _as_async_generator(self._response.aiter_bytes())
                 except Exception:  # noqa: BLE001 # Safe catch-all for cleanup logic
+                    try:
+                        await self._response.aread()
+                    except Exception:  # noqa: BLE001 S110 # Safe catch-all for cleanup logic
+                        pass
                     try:
                         await self._response.aclose()
                     except Exception:  # noqa: BLE001 S110 # Safe catch-all for cleanup logic

tests/test_litellm/passthrough/test_async_streaming_error_propagation.py

@@ -131,3 +131,30 @@ async def test_async_passthrough_wrapper_200_yields_chunks():
     assert len(chunks) == 1
     assert b"response.created" in chunks[0]
     mock_logging_obj.async_flush_passthrough_collected_chunks.assert_awaited_once()
+
+
+@pytest.mark.asyncio
+async def test_error_body_readable_after_failed_await():
+    """The upstream error body must stay readable so the proxy can map the real status and message."""
+    from litellm.passthrough.main import AsyncPassthroughStreamingResponse
+
+    error_body = b'{"message":"model not found"}'
+
+    async def byte_stream():
+        yield error_body
+
+    request = httpx.Request("POST", "https://bedrock.example.com/model/x/converse-stream")
+    response = httpx.Response(400, content=byte_stream(), request=request)
+
+    async def response_coro():
+        return response
+
+    with pytest.raises(httpx.HTTPStatusError) as exc_info:
+        await AsyncPassthroughStreamingResponse(
+            response=response_coro(),
+            litellm_logging_obj=_make_mock_logging_obj(),
+            provider_config=MagicMock(),
+        )
+
+    assert exc_info.value.response.status_code == 400
+    assert await exc_info.value.response.aread() == error_body