feat(config): add when_thinking_disabled support for model configs (#1970)

* feat(config): add when_thinking_disabled support for model configs

Allow users to explicitly configure what parameters are sent to the
model when thinking is disabled, via a new `when_thinking_disabled`
field in model config. This mirrors the existing `when_thinking_enabled`
pattern and takes full precedence over the hardcoded disable behavior
when set. Backwards compatible — existing configs work unchanged.

Closes #1675

* fix(config): address copilot review — gate when_thinking_disabled independently

- Switch truthiness check to `is not None` so empty dict overrides work
- Restructure disable path so when_thinking_disabled is gated independently
  of has_thinking_settings, allowing it to work without when_thinking_enabled
- Update test to reflect new behavior

backend/tests/test_model_factory.py

@@ -30,6 +30,7 @@ def _make_model(
     supports_thinking: bool = False,
     supports_reasoning_effort: bool = False,
     when_thinking_enabled: dict | None = None,
+    when_thinking_disabled: dict | None = None,
     thinking: dict | None = None,
     max_tokens: int | None = None,
 ) -> ModelConfig:
@@ -43,6 +44,7 @@ def _make_model(
         supports_thinking=supports_thinking,
         supports_reasoning_effort=supports_reasoning_effort,
         when_thinking_enabled=when_thinking_enabled,
+        when_thinking_disabled=when_thinking_disabled,
         thinking=thinking,
         supports_vision=False,
     )
@@ -244,6 +246,136 @@ def test_thinking_disabled_no_when_thinking_enabled_does_nothing(monkeypatch):
     assert captured.get("reasoning_effort") is None
 
 
+# ---------------------------------------------------------------------------
+# when_thinking_disabled config
+# ---------------------------------------------------------------------------
+
+
+def test_when_thinking_disabled_takes_precedence_over_hardcoded_disable(monkeypatch):
+    """When when_thinking_disabled is set, it takes full precedence over the
+    hardcoded disable logic (extra_body.thinking.type=disabled etc.)."""
+    wte = {"extra_body": {"thinking": {"type": "enabled", "budget_tokens": 10000}}}
+    wtd = {"extra_body": {"thinking": {"type": "disabled"}}, "reasoning_effort": "low"}
+    cfg = _make_app_config(
+        [
+            _make_model(
+                "custom-disable",
+                supports_thinking=True,
+                supports_reasoning_effort=True,
+                when_thinking_enabled=wte,
+                when_thinking_disabled=wtd,
+            )
+        ]
+    )
+    _patch_factory(monkeypatch, cfg)
+
+    captured: dict = {}
+
+    class CapturingModel(FakeChatModel):
+        def __init__(self, **kwargs):
+            captured.update(kwargs)
+            BaseChatModel.__init__(self, **kwargs)
+
+    monkeypatch.setattr(factory_module, "resolve_class", lambda path, base: CapturingModel)
+
+    factory_module.create_chat_model(name="custom-disable", thinking_enabled=False)
+
+    assert captured.get("extra_body") == {"thinking": {"type": "disabled"}}
+    # User overrode the hardcoded "minimal" with "low"
+    assert captured.get("reasoning_effort") == "low"
+
+
+def test_when_thinking_disabled_not_used_when_thinking_enabled(monkeypatch):
+    """when_thinking_disabled must have no effect when thinking_enabled=True."""
+    wte = {"extra_body": {"thinking": {"type": "enabled"}}}
+    wtd = {"extra_body": {"thinking": {"type": "disabled"}}}
+    cfg = _make_app_config(
+        [
+            _make_model(
+                "wtd-ignored",
+                supports_thinking=True,
+                when_thinking_enabled=wte,
+                when_thinking_disabled=wtd,
+            )
+        ]
+    )
+    _patch_factory(monkeypatch, cfg)
+
+    captured: dict = {}
+
+    class CapturingModel(FakeChatModel):
+        def __init__(self, **kwargs):
+            captured.update(kwargs)
+            BaseChatModel.__init__(self, **kwargs)
+
+    monkeypatch.setattr(factory_module, "resolve_class", lambda path, base: CapturingModel)
+
+    factory_module.create_chat_model(name="wtd-ignored", thinking_enabled=True)
+
+    # when_thinking_enabled should apply, NOT when_thinking_disabled
+    assert captured.get("extra_body") == {"thinking": {"type": "enabled"}}
+
+
+def test_when_thinking_disabled_without_when_thinking_enabled_still_applies(monkeypatch):
+    """when_thinking_disabled alone (no when_thinking_enabled) should still apply its settings."""
+    cfg = _make_app_config(
+        [
+            _make_model(
+                "wtd-only",
+                supports_thinking=True,
+                supports_reasoning_effort=True,
+                when_thinking_disabled={"reasoning_effort": "low"},
+            )
+        ]
+    )
+    _patch_factory(monkeypatch, cfg)
+
+    captured: dict = {}
+
+    class CapturingModel(FakeChatModel):
+        def __init__(self, **kwargs):
+            captured.update(kwargs)
+            BaseChatModel.__init__(self, **kwargs)
+
+    monkeypatch.setattr(factory_module, "resolve_class", lambda path, base: CapturingModel)
+
+    factory_module.create_chat_model(name="wtd-only", thinking_enabled=False)
+
+    # when_thinking_disabled is now gated independently of has_thinking_settings
+    assert captured.get("reasoning_effort") == "low"
+
+
+def test_when_thinking_disabled_excluded_from_model_dump(monkeypatch):
+    """when_thinking_disabled must not leak into the model constructor kwargs."""
+    wte = {"extra_body": {"thinking": {"type": "enabled"}}}
+    wtd = {"extra_body": {"thinking": {"type": "disabled"}}}
+    cfg = _make_app_config(
+        [
+            _make_model(
+                "no-leak-wtd",
+                supports_thinking=True,
+                when_thinking_enabled=wte,
+                when_thinking_disabled=wtd,
+            )
+        ]
+    )
+    _patch_factory(monkeypatch, cfg)
+
+    captured: dict = {}
+
+    class CapturingModel(FakeChatModel):
+        def __init__(self, **kwargs):
+            captured.update(kwargs)
+            BaseChatModel.__init__(self, **kwargs)
+
+    monkeypatch.setattr(factory_module, "resolve_class", lambda path, base: CapturingModel)
+
+    factory_module.create_chat_model(name="no-leak-wtd", thinking_enabled=True)
+
+    # when_thinking_disabled value must NOT appear as a raw key
+    assert "when_thinking_disabled" not in captured
+
+
 # ---------------------------------------------------------------------------
 # reasoning_effort stripping
 # ---------------------------------------------------------------------------