fix: support max reasoning and custom model selection for 1.0.0

tests/unit/admin_console/test_model_settings.py

@@ -125,16 +125,30 @@ def test_reasoning_retry_settings_roundtrip(isolated_settings):
     assert parse_llm_config().operator.reasoning_effort == "default"
 
 
-def test_responses_payload_has_reasoning_and_no_sdk_retry(monkeypatch):
+def test_max_reasoning_and_custom_model_survive_reload(isolated_settings):
+    result = service.save_model_settings(
+        service.ModelSettingsUpdate(model="custom/model-v2", reasoning_effort="max")
+    )
+    assert result["model"] == "custom/model-v2"
+    assert result["reasoning_effort"] == "max"
+    assert dotenv_values(isolated_settings)["OPENAI_REASONING_EFFORT"] == "max"
+    reloaded = Settings(_env_file=isolated_settings)
+    assert reloaded.OPENAI_REASONING_EFFORT == "max"
+    assert reloaded.OPENAI_MODEL == "custom/model-v2"
+    assert parse_llm_config().operator.reasoning_effort == "max"
+
+
+@pytest.mark.parametrize("effort", ["high", "max"])
+def test_responses_payload_has_reasoning_and_no_sdk_retry(monkeypatch, effort):
     from artemis.llm.router import ModelEndpoint, ModelFactory
     from unittest.mock import Mock
 
     factory = Mock()
     monkeypatch.setattr("langchain_openai.ChatOpenAI", factory)
     ModelFactory.create_model(
-        ModelEndpoint(api_key="test-only", reasoning_effort="high", max_retries=5)
+        ModelEndpoint(api_key="test-only", reasoning_effort=effort, max_retries=5)
     )
-    assert factory.call_args.kwargs["reasoning_effort"] == "high"
+    assert factory.call_args.kwargs["reasoning_effort"] == effort
     assert factory.call_args.kwargs["max_retries"] == 0
     assert factory.call_args.kwargs["use_responses_api"] is True
     ModelFactory.create_model(ModelEndpoint(api_key="test-only", reasoning_effort="default"))

tests/unit/test_openai_responses.py

@@ -85,6 +85,15 @@ async def test_text_and_image_use_responses_transport(transport_model):
     assert result.usage_metadata["total_tokens"] == 13
 
 
+@pytest.mark.asyncio
+async def test_max_reasoning_is_sent_unchanged_to_responses(transport_model):
+    model, requests, _ = transport_model
+    model.reasoning_effort = "max"
+    await model.ainvoke([HumanMessage(content="Check")])
+    assert requests[0].url.path == "/v1/responses"
+    assert json.loads(requests[0].content)["reasoning"]["effort"] == "max"
+
+
 @pytest.mark.asyncio
 async def test_function_call_roundtrip_uses_call_id(transport_model):
     model, requests, outputs = transport_model