chore(inputs): drop unused logger

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

@@ -58,7 +58,7 @@ if TYPE_CHECKING:
     from openai import AsyncOpenAI
     from openai.types.responses.response_prompt_param import ResponsePromptParam
 
-    from strix.config.settings import LlmSettings, ReasoningEffort, Settings
+    from strix.config.settings import ApiType, LlmSettings, ReasoningEffort, Settings
 
 
 logger = logging.getLogger(__name__)
@@ -649,14 +649,54 @@ 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 = llm.api_type
-    if api_type is None:
-        api_type = "chat_completions" if llm.api_base else "responses"
-
-    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"
+
+
+def resolve_api_type(model_name: str, settings: Settings) -> ApiType:
+    """The SDK-native OpenAI route for ``model_name``: Responses or chat completions.
+
+    An explicit ``STRIX_API_TYPE`` wins. Without a base URL the request goes to
+    OpenAI itself, where every model is served by the Responses API. With a base
+    URL the model decides, not the URL: a model whose catalog entry lists no
+    ``/v1/chat/completions`` endpoint cannot be reached there at all, and
+    api.openai.com behind a base URL is still OpenAI. Anything else is an
+    OpenAI-compatible gateway whose common denominator is chat completions.
+    """
+    if settings.llm.api_type is not None:
+        return settings.llm.api_type
+    api_base = (settings.llm.api_base or "").strip()
+    if not api_base or _is_openai_host(api_base):
+        return "responses"
+    endpoints = _catalog_supported_endpoints(model_name)
+    if endpoints and _CHAT_COMPLETIONS_ENDPOINT not in endpoints:
+        return "responses"
+    return "chat_completions"
+
+
+def _is_openai_host(api_base: str) -> bool:
+    from urllib.parse import urlsplit
+
+    return (urlsplit(api_base).hostname or "").lower() in _OPENAI_HOSTS
+
+
+def _catalog_supported_endpoints(model_name: str) -> list[str]:
+    entry = _catalog_entry(model_name)
+    endpoints = entry.get("supported_endpoints") if entry else None
+    return [str(e) for e in endpoints] if isinstance(endpoints, list) else []
+
+
 def _mirror_api_key_to_provider_env(model_name: str | None, api_key: str) -> None:
     if not model_name:
         return
@@ -859,10 +899,8 @@ def uses_chat_completions_tool_schema(model_name: str, settings: Settings) -> bo
     model = model_name.strip().lower()
     if "/" in model and not model.startswith("openai/"):
         return True
-    if settings.llm.api_type is not None:
-        return settings.llm.api_type == "chat_completions"
-    if settings.llm.api_base:
-        return True
+    if settings.llm.api_type is not None or settings.llm.api_base:
+        return resolve_api_type(model_name, settings) == "chat_completions"
     return not model_supports_reasoning(model_name)
 
 
@@ -879,17 +917,27 @@ def supports_strict_tool_schemas(model_name: str) -> bool:
 
 
 def model_supports_reasoning(model_name: str) -> bool:
-    import litellm
+    entry = _catalog_entry(model_name)
+    return bool(entry and entry.get("supports_reasoning"))
 
+
+def _bare_openai_name(model_name: str) -> str:
     name = model_name.strip().lower()
     for prefix in ("litellm/", "any-llm/", "openai/"):
         if name.startswith(prefix):
-            name = name[len(prefix) :]
-            break
+            return name[len(prefix) :]
+    return name
+
+
+def _catalog_entry(model_name: str) -> dict[str, Any] | None:
+    """LiteLLM's cost-map entry for the model, looked up as it would route it."""
+    import litellm
+
+    name = _bare_openai_name(model_name)
     entry = litellm.model_cost.get(name)
     if entry is None and "/" in name:
         entry = litellm.model_cost.get(name.rsplit("/", 1)[1])
-    return bool(entry and entry.get("supports_reasoning"))
+    return entry if isinstance(entry, dict) else None
 
 
 def is_known_openai_bare_model(model_name: str) -> bool:

strix/config/settings.py

@@ -4,7 +4,7 @@ from __future__ import annotations
 
 from typing import Literal
 
-from pydantic import AliasChoices, Field
+from pydantic import AliasChoices, Field, field_validator
 from pydantic_settings import BaseSettings, SettingsConfigDict
 
 
@@ -13,6 +13,12 @@ ApiType = Literal["responses", "chat_completions"]
 
 DEFAULT_MAX_TURNS = 500
 
+
+def _lowercase(value: object) -> object:
+    """Enum-like env values are matched case-insensitively (``None`` is ``none``)."""
+    return value.strip().lower() if isinstance(value, str) else value
+
+
 _BASE_CONFIG = SettingsConfigDict(
     case_sensitive=False,
     populate_by_name=True,
@@ -78,6 +84,11 @@ class LlmSettings(BaseSettings):
         alias="LLM_MAX_TOOL_CALLS_PER_TURN",
     )
 
+    @field_validator("api_type", "reasoning_effort", mode="before")
+    @classmethod
+    def _normalize_case(cls, value: object) -> object:
+        return _lowercase(value)
+
 
 class DedupeSettings(BaseSettings):
     model_config = _BASE_CONFIG
@@ -95,6 +106,11 @@ class DedupeSettings(BaseSettings):
         repr=False,
     )
 
+    @field_validator("reasoning_effort", mode="before")
+    @classmethod
+    def _normalize_case(cls, value: object) -> object:
+        return _lowercase(value)
+
 
 class ContextSettings(BaseSettings):
     """Context-window management: per-tool-output caps and history compaction."""

strix/core/inputs.py

@@ -267,9 +267,7 @@ def make_model_settings(
         and reasoning_effort != "none"
         and model_supports_reasoning(model_name)
     ):
-        model_settings = model_settings.resolve(
-            _reasoning_settings(reasoning_effort),
-        )
+        model_settings = model_settings.resolve(_reasoning_settings(reasoning_effort))
     if force_required_tool_choice and _accepts_required_tool_choice(model_name):
         model_settings = model_settings.resolve(ModelSettings(tool_choice="required"))
 

strix/core/runner.py

@@ -18,9 +18,10 @@ 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,
     supports_strict_tool_schemas,
     uses_chat_completions_tool_schema,
@@ -257,6 +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)

tests/test_inputs.py

@@ -430,3 +430,10 @@ def test_user_headers_override_openrouter_attribution() -> None:
     assert headers["X-Title"] == "Custom"
     assert headers["X-Tenant"] == "acme"
     assert headers["HTTP-Referer"] == "https://strix.ai"
+
+
+def test_reasoning_effort_sent_as_configured_and_none_omitted() -> None:
+    assert make_model_settings("none", model_name="gpt-5.6-sol").reasoning is None
+    settings = make_model_settings("high", model_name="gpt-daybreak-blue-latest")
+    assert settings.reasoning is not None
+    assert settings.reasoning.effort == "high"

tests/test_models.py

@@ -10,12 +10,14 @@ 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,
     _TurnGuardModel,
     configure_sdk_model_defaults,
     request_timeout_extra_args,
+    resolve_api_type,
     routes_through_litellm,
     supports_strict_tool_schemas,
     uses_chat_completions_tool_schema,
@@ -138,3 +140,80 @@ def test_api_type_overrides_the_api_base_route(
     while isinstance(model, _NonStreamingModel | _TurnGuardModel | RequestLoggingModel):
         model = model._inner
     assert isinstance(model, expected)
+
+
+def _settings(monkeypatch: pytest.MonkeyPatch, model: str, api_base: str | None) -> Settings:
+    monkeypatch.setenv("STRIX_LLM", model)
+    monkeypatch.delenv("STRIX_API_TYPE", raising=False)
+    for name in ("LLM_API_BASE", "OPENAI_API_BASE", "OPENAI_BASE_URL"):
+        monkeypatch.delenv(name, raising=False)
+    if api_base is not None:
+        monkeypatch.setenv("LLM_API_BASE", api_base)
+    return Settings()
+
+
+def test_resolve_api_type_without_base_url_is_responses(monkeypatch: pytest.MonkeyPatch) -> None:
+    assert resolve_api_type("gpt-5", _settings(monkeypatch, "gpt-5", None)) == "responses"
+    assert resolve_api_type("gpt-4o", _settings(monkeypatch, "gpt-4o", None)) == "responses"
+
+
+def test_resolve_api_type_gateway_defaults_to_chat_completions(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    settings = _settings(monkeypatch, "gpt-5.6-sol", "https://gateway.example/v1")
+    assert resolve_api_type("gpt-5.6-sol", settings) == "chat_completions"
+    assert resolve_api_type("my-private-model", settings) == "chat_completions"
+
+
+def test_resolve_api_type_responses_only_model_ignores_base_url(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """A model LiteLLM lists on /v1/responses alone cannot be served by chat completions."""
+    settings = _settings(monkeypatch, "gpt-daybreak-blue-latest", "https://gateway.example/v1")
+    assert resolve_api_type("gpt-daybreak-blue-latest", settings) == "responses"
+    assert resolve_api_type("openai/gpt-daybreak-blue-latest", settings) == "responses"
+    assert uses_chat_completions_tool_schema("gpt-daybreak-blue-latest", settings) is False
+
+
+def test_resolve_api_type_openai_host_behind_base_url_is_responses(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    settings = _settings(monkeypatch, "gpt-5", "https://api.openai.com/v1")
+    assert resolve_api_type("gpt-5", settings) == "responses"
+    assert uses_chat_completions_tool_schema("gpt-5", settings) is False
+
+
+def test_resolve_api_type_explicit_override_wins(monkeypatch: pytest.MonkeyPatch) -> None:
+    _settings(monkeypatch, "gpt-daybreak-blue-latest", "https://gateway.example/v1")
+    monkeypatch.setenv("STRIX_API_TYPE", "chat_completions")
+    assert resolve_api_type("gpt-daybreak-blue-latest", Settings()) == "chat_completions"
+    monkeypatch.setenv("STRIX_API_TYPE", "Responses")
+    assert resolve_api_type("gpt-5", Settings()) == "responses"
+
+
+@pytest.mark.parametrize(
+    ("raw", "expected"),
+    [("None", "none"), ("HIGH", "high"), (" xhigh ", "xhigh"), ("Max", "max")],
+)
+def test_reasoning_effort_is_case_insensitive(
+    monkeypatch: pytest.MonkeyPatch, raw: str, expected: str
+) -> None:
+    monkeypatch.setenv("STRIX_REASONING_EFFORT", raw)
+    monkeypatch.setenv("STRIX_DEDUPE_REASONING_EFFORT", raw)
+    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"]