Skip to content

Drop in a screenshot and convert it to clean code (HTML/Tailwind/React/Vue)

Python

set prompt cache retention to 24h for GPT 5.4

Abi Raja5fb885f2f287
diff --git a/backend/agent/providers/openai.py b/backend/agent/providers/openai.py
index 818033a9..571ef411 100644
--- a/backend/agent/providers/openai.py
+++ b/backend/agent/providers/openai.py
@@ -455,8 +455,9 @@ class OpenAIProviderSession(ProviderSession):
)
async def stream_turn(self, on_event: EventSink) -> ProviderTurn:
+ model_name = get_openai_api_name(self._model)
params: Dict[str, Any] = {
- "model": get_openai_api_name(self._model),
+ "model": model_name,
"input": self._input_items,
"tools": self._tools,
"tool_choice": "auto",
@@ -464,6 +465,8 @@ class OpenAIProviderSession(ProviderSession):
"max_output_tokens": 50000,
"prompt_cache_key": self._prompt_cache_key,
}
+ if model_name == "gpt-5.4-2026-03-05":
+ params["prompt_cache_retention"] = "24h"
reasoning_effort = get_openai_reasoning_effort(self._model)
if reasoning_effort:
params["reasoning"] = {"effort": reasoning_effort, "summary": "auto"}
diff --git a/backend/tests/test_openai_provider_session.py b/backend/tests/test_openai_provider_session.py
index ad45fddb..6117e2aa 100644
--- a/backend/tests/test_openai_provider_session.py
+++ b/backend/tests/test_openai_provider_session.py
@@ -111,6 +111,8 @@ async def test_openai_provider_session_reuses_prompt_cache_key_across_turns() ->
assert first_call["prompt_cache_key"] == second_call["prompt_cache_key"]
assert isinstance(first_call["prompt_cache_key"], str)
assert str(first_call["prompt_cache_key"]).startswith("s2c-openai-session-v1-")
+ assert "prompt_cache_retention" not in first_call
+ assert "prompt_cache_retention" not in second_call
assert isinstance(first_input, list)
assert isinstance(second_input, list)
assert len(second_input) > len(first_input)
@@ -168,6 +170,7 @@ async def test_openai_provider_session_uses_gpt_5_4_none_reasoning_effort() -> N
first_call = client.responses.calls[0]
assert first_call["model"] == "gpt-5.4-2026-03-05"
+ assert first_call["prompt_cache_retention"] == "24h"
assert first_call["reasoning"] == {"effort": "none", "summary": "auto"}
@@ -186,4 +189,5 @@ async def test_openai_provider_session_uses_gpt_5_4_high_reasoning_effort() -> N
first_call = client.responses.calls[0]
assert first_call["model"] == "gpt-5.4-2026-03-05"
+ assert first_call["prompt_cache_retention"] == "24h"
assert first_call["reasoning"] == {"effort": "high", "summary": "auto"}