merge(dev): integrate latest upstream dev (PR #6498 omit temperature from subscription requests)

src/llm_core.py

@@ -1556,8 +1556,9 @@ def _build_chatgpt_responses_payload(
         "stream": stream,
         "store": False,
     }
-    if not _restricts_temperature(model):
-        payload["temperature"] = temperature
+    # ChatGPT plan Responses requests reject temperature for every model.
+    # The common caller still passes its preset value, but this provider must
+    # omit the field even for model names outside the known reasoning families.
     # ChatGPT Subscription Codex API does not support max_output_tokens —
     # passing it returns HTTP 400 "Unsupported parameter: max_output_tokens".
     # Do not include it in the payload.

tests/test_llm_core_temperature_reasoning.py

@@ -156,3 +156,18 @@ def test_chatgpt_subscription_payload_omits_max_output_tokens_when_zero():
     )
 
     assert "max_output_tokens" not in payload
+
+
+@pytest.mark.parametrize("model", ["gpt-5.6-luna", "gpt-6-luna", "gpt-6-sol", "gpt-6.1-sol", "gpt-4.1"])
+def test_chatgpt_subscription_payload_omits_temperature_for_every_model(model):
+    payload = llm_core._build_chatgpt_responses_payload(
+        model,
+        [{"role": "user", "content": "Say OK"}],
+        temperature=0.2,
+        max_tokens=37,
+        stream=True,
+    )
+
+    assert "temperature" not in payload
+    assert payload["stream"] is True
+    assert payload["store"] is False