diff --git a/src/agents/run_internal/session_persistence.py b/src/agents/run_internal/session_persistence.py
index 187924e895..780db80b2f 100644
--- a/src/agents/run_internal/session_persistence.py
+++ b/src/agents/run_internal/session_persistence.py
@@ -333,6 +333,14 @@ async def save_result_to_session(
             serialized_to_save_counts[serialized] -= 1
             saved_run_items_count += 1
 
+    # Drop items the Conversations API cannot accept (counted above so the
+    # persisted-item counter advances correctly on the next retry/call).
+    if is_openai_conversation_session:
+        items_to_save = [
+            item for item in items_to_save
+            if not _is_unpersistable_for_openai_conversation(item)
+        ]
+
     if len(items_to_save) == 0:
         if run_state:
             run_state._current_turn_persisted_item_count = already_persisted + saved_run_items_count
@@ -571,6 +579,19 @@ def _sanitize_openai_conversation_item(item: TResponseInputItem) -> TResponseInp
     return item
 
 
+def _is_unpersistable_for_openai_conversation(item: TResponseInputItem) -> bool:
+    """Return True for items the OpenAI Conversations API cannot accept.
+
+    The Conversations API rejects reasoning items with an empty summary list.
+    Callers should drop these items before calling ``session.add_items`` while
+    still counting them in the persisted-item counter so the retry logic
+    advances past them correctly.
+    """
+    if not isinstance(item, dict):
+        return False
+    return item.get("type") == "reasoning" and item.get("summary") == []
+
+
 def _sanitize_openai_conversation_history_items_for_model_input(
     items: Sequence[TResponseInputItem],
     history_indexes: set[int],
diff --git a/tests/test_agent_runner.py b/tests/test_agent_runner.py
index 284927a984..f1eaaffb76 100644
--- a/tests/test_agent_runner.py
+++ b/tests/test_agent_runner.py
@@ -2796,6 +2796,105 @@ async def test_save_result_to_session_omits_reasoning_ids_when_policy_is_omit()
     assert "id" not in saved_reasoning
 
 
+@pytest.mark.asyncio
+async def test_save_result_to_session_drops_empty_reasoning_from_openai_conversations() -> None:
+    """OpenAIConversationsSession must not receive reasoning items with summary=[]."""
+
+    class DummyOpenAIConversationsSession(OpenAIConversationsSession):
+        def __init__(self) -> None:
+            self.saved_items: list[TResponseInputItem] = []
+
+        async def _get_session_id(self) -> str:
+            return "conv_test"
+
+        async def add_items(self, items: list[TResponseInputItem]) -> None:
+            self.saved_items.extend(items)
+
+        async def get_items(self, limit: int | None = None) -> list[TResponseInputItem]:
+            return []
+
+        async def pop_item(self) -> TResponseInputItem | None:
+            return None
+
+        async def clear_session(self) -> None:
+            return None
+
+    session = DummyOpenAIConversationsSession()
+    agent = Agent(name="agent", model=FakeModel())
+    run_state: RunState[Any] = RunState(
+        context=RunContextWrapper(context={}),
+        original_input="input",
+        starting_agent=agent,
+        max_turns=1,
+    )
+
+    empty_reasoning = ReasoningItem(
+        agent=agent,
+        raw_item=ResponseReasoningItem(type="reasoning", id="rs_empty", summary=[]),
+    )
+
+    saved_count = await save_result_to_session(
+        session,
+        [],
+        cast(list[RunItem], [empty_reasoning]),
+        run_state,
+    )
+
+    # The item is counted (so the retry counter advances) but not sent to add_items.
+    assert saved_count == 1
+    assert run_state._current_turn_persisted_item_count == 1
+    assert len(session.saved_items) == 0
+
+
+@pytest.mark.asyncio
+async def test_save_result_to_session_keeps_nonempty_reasoning_in_openai_conversations() -> None:
+    """Non-empty reasoning items must still be persisted to OpenAIConversationsSession."""
+
+    class DummyOpenAIConversationsSession(OpenAIConversationsSession):
+        def __init__(self) -> None:
+            self.saved_items: list[TResponseInputItem] = []
+
+        async def _get_session_id(self) -> str:
+            return "conv_test"
+
+        async def add_items(self, items: list[TResponseInputItem]) -> None:
+            self.saved_items.extend(items)
+
+        async def get_items(self, limit: int | None = None) -> list[TResponseInputItem]:
+            return []
+
+        async def pop_item(self) -> TResponseInputItem | None:
+            return None
+
+        async def clear_session(self) -> None:
+            return None
+
+    session = DummyOpenAIConversationsSession()
+    agent = Agent(name="agent", model=FakeModel())
+
+    nonempty_reasoning = ReasoningItem(
+        agent=agent,
+        raw_item=ResponseReasoningItem(
+            type="reasoning",
+            id="rs_full",
+            summary=[Summary(type="summary_text", text="I thought about it")],
+        ),
+    )
+
+    saved_count = await save_result_to_session(
+        session,
+        [],
+        cast(list[RunItem], [nonempty_reasoning]),
+        None,
+    )
+
+    assert saved_count == 1
+    assert len(session.saved_items) == 1
+    saved = cast(dict[str, Any], session.saved_items[0])
+    assert saved.get("type") == "reasoning"
+    assert saved.get("summary") != []
+
+
 @pytest.mark.asyncio
 async def test_save_result_to_session_keeps_tool_call_payload_api_safe() -> None:
     session = SimpleListSession()