fix(responses): merge text-only items with output items in SSE recovery

When recovering output from raw SSE, OUTPUT_ITEM_DONE and OUTPUT_TEXT_DONE
events were treated as mutually exclusive fallbacks. If a stream emitted
OUTPUT_ITEM_DONE for some output indices and only OUTPUT_TEXT_DONE for
others, the text-only items at the missing indices were silently dropped.

Merge both dicts before returning, with OUTPUT_ITEM_DONE entries taking
precedence at any shared index (preserving the existing behavior covered
by test_transform_response_preserves_output_item_when_text_done_arrives_later).

Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>

litellm/completion_extras/litellm_responses_transformation/transformation.py

@@ -735,11 +735,16 @@ class LiteLLMResponsesTransformationHandler(CompletionTransformationBridge):
                     recovered_text_only_items=recovered_text_only_items,
                 )
 
-        if recovered_output_items:
-            return [item for _, item in sorted(recovered_output_items.items())]
+        # Merge text-only items into the recovered output items. Real
+        # OUTPUT_ITEM_DONE events take precedence at any given output_index,
+        # but text-only items at indices without a matching OUTPUT_ITEM_DONE
+        # must still be preserved (e.g. multi-output responses where some
+        # indices only emitted OUTPUT_TEXT_DONE).
+        merged_items: Dict[int, Dict[str, Any]] = {**recovered_text_only_items}
+        merged_items.update(recovered_output_items)
 
-        if recovered_text_only_items:
-            return [item for _, item in sorted(recovered_text_only_items.items())]
+        if merged_items:
+            return [item for _, item in sorted(merged_items.items())]
 
         return []
 

tests/test_litellm/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py

@@ -743,6 +743,36 @@ def test_transform_response_preserves_output_item_when_text_done_arrives_later()
     assert result.choices[0].message.content == "Complete output item text"
 
 
+def test_recover_output_items_merges_text_only_items_at_distinct_indices():
+    """When OUTPUT_ITEM_DONE covers some indices and OUTPUT_TEXT_DONE covers
+    others, both must be preserved instead of treating them as mutually
+    exclusive fallbacks."""
+    from litellm.completion_extras.litellm_responses_transformation.transformation import (
+        LiteLLMResponsesTransformationHandler,
+    )
+
+    raw_sse = "\n".join(
+        [
+            'data: {"type":"response.output_item.done","output_index":0,"item":{"type":"message","id":"msg_item_0","role":"assistant","status":"completed","content":[{"type":"output_text","text":"From OUTPUT_ITEM_DONE","annotations":[]}]}}',
+            'data: {"type":"response.output_text.done","output_index":1,"content_index":0,"item_id":"msg_text_1","text":"From OUTPUT_TEXT_DONE only"}',
+            "data: [DONE]",
+            "",
+        ]
+    )
+
+    recovered = (
+        LiteLLMResponsesTransformationHandler._recover_output_items_from_raw_sse(
+            raw_sse
+        )
+    )
+
+    assert len(recovered) == 2
+    assert recovered[0]["id"] == "msg_item_0"
+    assert recovered[0]["content"][0]["text"] == "From OUTPUT_ITEM_DONE"
+    assert recovered[1]["id"] == "msg_text_1"
+    assert recovered[1]["content"][0]["text"] == "From OUTPUT_TEXT_DONE only"
+
+
 def test_transform_response_prefers_completed_output_from_raw_sse():
     from litellm.completion_extras.litellm_responses_transformation.transformation import (
         LiteLLMResponsesTransformationHandler,