diff --git a/tests/lib/streaming/agents/test_streams.py b/tests/lib/streaming/agents/test_streams.py
index 0e77bcabff..b1495bfc65 100644
--- a/tests/lib/streaming/agents/test_streams.py
+++ b/tests/lib/streaming/agents/test_streams.py
@@ -481,7 +481,8 @@ def handler(arguments: dict[str, object]) -> str:
 
     repeated_call = {"event_id": "original-call", **call()}
     payloads = [turn_event("created"), repeated_call]
-    for index in range(5000):
+    delta_count = 5000
+    for index in range(delta_count):
         delta_event: dict[str, object] = {
             "type": "agent.session.turn.output_text.delta",
             "event_id": f"delta-{index}",
@@ -492,10 +493,12 @@ def handler(arguments: dict[str, object]) -> str:
             "content_index": 0,
             "delta": str(index),
         }
-        payloads.extend([delta_event, delta_event])
+        payloads.append(delta_event)
+    payloads.append(payloads[-1])  # Repeat the last delta after sustained cache eviction.
     payloads.extend([repeated_call, turn_event("completed"), idle()])
     server.body = EventBody(payloads)
     deltas: list[str] = []
+    tool_event_deliveries = 0
     if isinstance(sdk, AsyncOpenAI):
         async with sdk.beta.agents.sessions.stream(
             "session_test", input="Hello", tool_handlers={"search": handler}
@@ -503,6 +506,8 @@ def handler(arguments: dict[str, object]) -> str:
             async for event in stream:
                 if event.type == "agent.session.turn.output_text.delta":
                     deltas.append(event.delta)
+                elif event.event_id == repeated_call["event_id"]:
+                    tool_event_deliveries += 1
             assert len(stream._state.event_ids) <= 1024
             assert len(stream._state.recent_events) <= 1024
     else:
@@ -512,9 +517,12 @@ def handler(arguments: dict[str, object]) -> str:
             for event in sync_stream:
                 if event.type == "agent.session.turn.output_text.delta":
                     deltas.append(event.delta)
+                elif event.event_id == repeated_call["event_id"]:
+                    tool_event_deliveries += 1
             assert len(sync_stream._state.event_ids) <= 1024
             assert len(sync_stream._state.recent_events) <= 1024
-    assert deltas == [str(index) for index in range(5000)]
+    assert deltas == [str(index) for index in range(delta_count)]
+    assert tool_event_deliveries == 2
     assert calls == [{"query": "test"}]
     assert len(server.inputs()) == 2
     assert server.body.closed