Add per-model thinking toggles

tests/api/test_api.py

@@ -100,7 +100,9 @@ def test_model_mapping(client: TestClient):
     client.post("/v1/messages", json=payload_haiku)
     assert len(_stream_response_calls) == 1
     args = _stream_response_calls[0][0]
+    kwargs = _stream_response_calls[0][1]
     assert args[0].model != "claude-3-haiku-20240307"
+    assert kwargs["thinking_enabled"] is True
 
 
 def test_error_fallbacks(client: TestClient):

tests/api/test_dependencies.py

@@ -39,7 +39,7 @@ def _make_mock_settings(**overrides):
     mock.http_read_timeout = 300.0
     mock.http_write_timeout = 10.0
     mock.http_connect_timeout = 2.0
-    mock.enable_thinking = True
+    mock.enable_model_thinking = True
     for key, value in overrides.items():
         setattr(mock, key, value)
     return mock
@@ -159,12 +159,12 @@ async def test_get_provider_deepseek_uses_fixed_base_url():
 
 
 @pytest.mark.asyncio
-async def test_get_provider_deepseek_passes_enable_thinking():
-    """DeepSeek provider receives the global thinking toggle."""
+async def test_get_provider_deepseek_passes_enable_model_thinking():
+    """DeepSeek provider receives the fallback thinking toggle."""
     with patch("api.dependencies.get_settings") as mock_settings:
         mock_settings.return_value = _make_mock_settings(
             provider_type="deepseek",
-            enable_thinking=False,
+            enable_model_thinking=False,
         )
 
         provider = get_provider()

tests/api/test_model_router.py

@@ -14,6 +14,10 @@ def settings():
     settings.model_opus = None
     settings.model_sonnet = None
     settings.model_haiku = None
+    settings.enable_model_thinking = True
+    settings.enable_opus_thinking = None
+    settings.enable_sonnet_thinking = None
+    settings.enable_haiku_thinking = None
     return settings
 
 
@@ -24,6 +28,7 @@ def test_model_router_resolves_default_model(settings):
     assert resolved.provider_id == "nvidia_nim"
     assert resolved.provider_model == "fallback-model"
     assert resolved.provider_model_ref == "nvidia_nim/fallback-model"
+    assert resolved.thinking_enabled is True
 
 
 def test_model_router_applies_opus_override(settings):
@@ -39,9 +44,23 @@ def test_model_router_applies_opus_override(settings):
     assert routed.request.model == "deepseek/deepseek-r1"
     assert routed.resolved.provider_model_ref == "open_router/deepseek/deepseek-r1"
     assert routed.resolved.original_model == "claude-opus-4-20250514"
+    assert routed.resolved.thinking_enabled is True
     assert request.model == "claude-opus-4-20250514"
 
 
+def test_model_router_resolves_per_model_thinking(settings):
+    settings.enable_model_thinking = False
+    settings.enable_opus_thinking = True
+    settings.enable_haiku_thinking = False
+
+    router = ModelRouter(settings)
+
+    assert router.resolve("claude-opus-4-20250514").thinking_enabled is True
+    assert router.resolve("claude-sonnet-4-20250514").thinking_enabled is False
+    assert router.resolve("claude-3-haiku-20240307").thinking_enabled is False
+    assert router.resolve("claude-2.1").thinking_enabled is False
+
+
 def test_model_router_applies_haiku_override(settings):
     settings.model_haiku = "lmstudio/qwen2.5-7b"
 

tests/config/test_config.py

@@ -29,7 +29,7 @@ class TestSettings:
         assert isinstance(settings.provider_rate_window, int)
         assert isinstance(settings.nim.temperature, float)
         assert isinstance(settings.fast_prefix_detection, bool)
-        assert isinstance(settings.enable_thinking, bool)
+        assert isinstance(settings.enable_model_thinking, bool)
         assert settings.http_read_timeout == 120.0
 
     def test_get_settings_cached(self):
@@ -110,13 +110,48 @@ class TestSettings:
         settings = Settings()
         assert settings.http_connect_timeout == 5.0
 
-    def test_enable_thinking_from_env(self, monkeypatch):
-        """ENABLE_THINKING env var is loaded into settings."""
+    def test_enable_model_thinking_from_env(self, monkeypatch):
+        """ENABLE_MODEL_THINKING env var is loaded into settings."""
         from config.settings import Settings
 
-        monkeypatch.setenv("ENABLE_THINKING", "false")
+        monkeypatch.setenv("ENABLE_MODEL_THINKING", "false")
         settings = Settings()
-        assert settings.enable_thinking is False
+        assert settings.enable_model_thinking is False
+
+    def test_per_model_thinking_from_env(self, monkeypatch):
+        """Per-model thinking env vars are loaded into settings."""
+        from config.settings import Settings
+
+        monkeypatch.setenv("ENABLE_OPUS_THINKING", "true")
+        monkeypatch.setenv("ENABLE_SONNET_THINKING", "false")
+        monkeypatch.setenv("ENABLE_HAIKU_THINKING", "false")
+        settings = Settings()
+        assert settings.enable_opus_thinking is True
+        assert settings.enable_sonnet_thinking is False
+        assert settings.enable_haiku_thinking is False
+
+    def test_empty_per_model_thinking_inherits_model_default(self, monkeypatch):
+        """Blank per-model thinking env vars are treated as unset."""
+        from config.settings import Settings
+
+        monkeypatch.setenv("ENABLE_MODEL_THINKING", "false")
+        monkeypatch.setenv("ENABLE_OPUS_THINKING", "")
+        settings = Settings()
+        assert settings.enable_opus_thinking is None
+        assert settings.resolve_thinking("claude-opus-4-20250514") is False
+
+    def test_resolve_thinking_uses_model_tiers(self, monkeypatch):
+        """resolve_thinking applies tier override then fallback."""
+        from config.settings import Settings
+
+        monkeypatch.setenv("ENABLE_MODEL_THINKING", "false")
+        monkeypatch.setenv("ENABLE_OPUS_THINKING", "true")
+        monkeypatch.setenv("ENABLE_HAIKU_THINKING", "false")
+        settings = Settings()
+        assert settings.resolve_thinking("claude-opus-4-20250514") is True
+        assert settings.resolve_thinking("claude-sonnet-4-20250514") is False
+        assert settings.resolve_thinking("claude-haiku-4-20250514") is False
+        assert settings.resolve_thinking("unknown-model") is False
 
     def test_anthropic_auth_token_from_env_without_dotenv_key(self, monkeypatch):
         """ANTHROPIC_AUTH_TOKEN env var is loaded when dotenv does not define it."""
@@ -166,7 +201,15 @@ class TestSettings:
         from config.settings import Settings
 
         monkeypatch.setenv("NIM_ENABLE_THINKING", "false")
-        with pytest.raises(ValidationError, match="Rename it to ENABLE_THINKING"):
+        with pytest.raises(ValidationError, match="ENABLE_MODEL_THINKING"):
+            Settings()
+
+    def test_removed_enable_thinking_raises(self, monkeypatch):
+        """ENABLE_THINKING now fails fast with a migration message."""
+        from config.settings import Settings
+
+        monkeypatch.setenv("ENABLE_THINKING", "false")
+        with pytest.raises(ValidationError, match="ENABLE_MODEL_THINKING"):
             Settings()
 
 

tests/providers/test_registry.py

@@ -38,7 +38,7 @@ def _make_settings(**overrides):
     mock.http_read_timeout = 300.0
     mock.http_write_timeout = 10.0
     mock.http_connect_timeout = 2.0
-    mock.enable_thinking = True
+    mock.enable_model_thinking = True
     mock.nim = NimSettings()
     for key, value in overrides.items():
         setattr(mock, key, value)