fix(runner): pick the SDK route from the resolved model override

configure_sdk_model_defaults only sees STRIX_LLM; a model= override to
run_strix_scan re-applies the route for the model that actually runs.

strix/config/models.py

@@ -649,12 +649,17 @@ def configure_sdk_model_defaults(settings: Settings) -> None:
     if llm.api_base:
         os.environ["OPENAI_BASE_URL"] = llm.api_base
         _configure_litellm_default("api_base", llm.api_base)
-    api_type = resolve_api_type(llm.model or "", settings)
-    logger.info("OpenAI API route: %s", api_type)
-    set_default_openai_api(api_type)
+    configure_sdk_api_route(llm.model or "", settings)
     _configure_extra_headers(llm)
 
 
+def configure_sdk_api_route(model_name: str, settings: Settings) -> None:
+    """Point SDK-native OpenAI requests for ``model_name`` at Responses or chat completions."""
+    api_type = resolve_api_type(model_name, settings)
+    logger.info("OpenAI API route for %s: %s", model_name, api_type)
+    set_default_openai_api(api_type)
+
+
 _OPENAI_HOSTS = frozenset({"api.openai.com"})
 _CHAT_COMPLETIONS_ENDPOINT = "/v1/chat/completions"
 

strix/core/runner.py

@@ -18,11 +18,11 @@ from openai import RateLimitError
 
 from strix.agents.factory import build_strix_agent, make_child_factory
 from strix.agents.prompt import render_scope_prompt, render_system_prompt
-from strix.config import load_settings
+from strix.config import codex, load_settings
 from strix.config.models import (
     StrixProvider,
+    configure_sdk_api_route,
     configure_sdk_model_defaults,
-    resolve_api_type,
     supports_strict_tool_schemas,
     uses_chat_completions_tool_schema,
 )
@@ -258,6 +258,10 @@ async def run_strix_scan(
         raise RuntimeError(
             "No LLM model configured. Set STRIX_LLM env or pass model= to run_strix_scan().",
         )
+    if resolved_model != (settings.llm.model or "").strip() and not codex.subscription_model(
+        resolved_model
+    ):
+        configure_sdk_api_route(resolved_model, settings)
     logger.info("LLM model resolved: %s", resolved_model)
     chat_completions_tools = uses_chat_completions_tool_schema(resolved_model, settings)
     strict_tool_schemas = supports_strict_tool_schemas(resolved_model)
@@ -373,7 +377,7 @@ async def run_strix_scan(
             request_timeout=settings.llm.timeout,
             prompt_cache=settings.llm.prompt_cache,
             extra_headers=settings.llm.extra_headers,
-            api_type=resolve_api_type(resolved_model, settings),
+            api_type="chat_completions" if chat_completions_tools else "responses",
         )
         run_config = RunConfig(
             model=resolved_model,

tests/test_models.py

@@ -10,6 +10,7 @@ from agents.models import _openai_shared
 from agents.models.openai_chatcompletions import OpenAIChatCompletionsModel
 from agents.models.openai_responses import OpenAIResponsesModel
 
+from strix.config import models
 from strix.config.models import (
     StrixProvider,
     _NonStreamingModel,
@@ -202,3 +203,17 @@ def test_reasoning_effort_is_case_insensitive(
     settings = Settings()
     assert settings.llm.reasoning_effort == expected
     assert settings.dedupe.reasoning_effort == expected
+
+
+def test_configure_sdk_api_route_follows_the_given_model(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """A ``model=`` override picks its own route, not ``STRIX_LLM``'s."""
+    routes: list[str] = []
+    monkeypatch.setattr(models, "set_default_openai_api", routes.append)
+    settings = _settings(monkeypatch, "gpt-5", "https://gateway.example/v1")
+
+    models.configure_sdk_api_route("gpt-5", settings)
+    models.configure_sdk_api_route("gpt-daybreak-blue-latest", settings)
+
+    assert routes == ["chat_completions", "responses"]