Enable minimal thinking budget by default for Anthropic integration (#167593)

homeassistant/components/anthropic/const.py

@@ -36,13 +36,15 @@ class PromptCaching(StrEnum):
     AUTOMATIC = "automatic"
 
 
+MIN_THINKING_BUDGET = 1024
+
 DEFAULT = {
     CONF_CHAT_MODEL: "claude-haiku-4-5",
     CONF_CODE_EXECUTION: False,
     CONF_MAX_TOKENS: 3000,
     CONF_PROMPT_CACHING: PromptCaching.PROMPT.value,
     CONF_TEMPERATURE: 1.0,
-    CONF_THINKING_BUDGET: 0,
+    CONF_THINKING_BUDGET: MIN_THINKING_BUDGET,
     CONF_THINKING_EFFORT: "low",
     CONF_TOOL_SEARCH: False,
     CONF_WEB_SEARCH: False,
@@ -50,8 +52,6 @@ DEFAULT = {
     CONF_WEB_SEARCH_MAX_USES: 5,
 }
 
-MIN_THINKING_BUDGET = 1024
-
 NON_THINKING_MODELS = [
     "claude-3-haiku",
 ]

tests/components/anthropic/snapshots/test_ai_task.ambr

@@ -47,9 +47,9 @@
         'type': 'text',
       }),
     ]),
-    'temperature': 1.0,
     'thinking': dict({
-      'type': 'disabled',
+      'budget_tokens': 1024,
+      'type': 'enabled',
     }),
   })
 # ---

tests/components/anthropic/snapshots/test_conversation.ambr

@@ -195,7 +195,7 @@
     }),
   ])
 # ---
-# name: test_disabled_thinking
+# name: test_disabled_thinking[subentry_data0]
   list([
     dict({
       'content': '''
@@ -224,7 +224,72 @@
     }),
   ])
 # ---
-# name: test_disabled_thinking.1
+# name: test_disabled_thinking[subentry_data0].1
+  dict({
+    'container': None,
+    'max_tokens': 3000,
+    'messages': list([
+      dict({
+        'content': 'hello',
+        'role': 'user',
+      }),
+      dict({
+        'content': 'Hello, how can I help you today?',
+        'role': 'assistant',
+      }),
+    ]),
+    'model': 'claude-haiku-4-5',
+    'stream': True,
+    'system': list([
+      dict({
+        'cache_control': dict({
+          'type': 'ephemeral',
+        }),
+        'text': '''
+          You are a voice assistant for Home Assistant.
+          Answer questions about the world truthfully.
+          Answer in plain text. Keep it simple and to the point.
+          Only if the user wants to control a device, tell them to expose entities to their voice assistant in Home Assistant.
+        ''',
+        'type': 'text',
+      }),
+    ]),
+    'temperature': 1.0,
+    'thinking': dict({
+      'type': 'disabled',
+    }),
+  })
+# ---
+# name: test_disabled_thinking[subentry_data1]
+  list([
+    dict({
+      'content': '''
+        You are a voice assistant for Home Assistant.
+        Answer questions about the world truthfully.
+        Answer in plain text. Keep it simple and to the point.
+        Only if the user wants to control a device, tell them to expose entities to their voice assistant in Home Assistant.
+      ''',
+      'created': HAFakeDatetime(2024, 5, 24, 12, 0, tzinfo=datetime.timezone.utc),
+      'role': 'system',
+    }),
+    dict({
+      'attachments': None,
+      'content': 'hello',
+      'created': HAFakeDatetime(2024, 5, 24, 12, 0, tzinfo=datetime.timezone.utc),
+      'role': 'user',
+    }),
+    dict({
+      'agent_id': 'conversation.claude_conversation',
+      'content': 'Hello, how can I help you today?',
+      'created': HAFakeDatetime(2024, 5, 24, 12, 0, tzinfo=datetime.timezone.utc),
+      'native': None,
+      'role': 'assistant',
+      'thinking_content': None,
+      'tool_calls': None,
+    }),
+  ])
+# ---
+# name: test_disabled_thinking[subentry_data1].1
   dict({
     'container': None,
     'max_tokens': 3000,

tests/components/anthropic/test_ai_task.py

@@ -10,7 +10,10 @@ from syrupy.assertion import SnapshotAssertion
 import voluptuous as vol
 
 from homeassistant.components import ai_task, media_source
-from homeassistant.components.anthropic.const import CONF_CHAT_MODEL
+from homeassistant.components.anthropic.const import (
+    CONF_CHAT_MODEL,
+    CONF_THINKING_BUDGET,
+)
 from homeassistant.core import HomeAssistant
 from homeassistant.exceptions import HomeAssistantError
 from homeassistant.helpers import entity_registry as er, selector
@@ -127,6 +130,7 @@ async def test_generate_structured_data_legacy(
             subentry,
             data={
                 CONF_CHAT_MODEL: "claude-sonnet-4-0",
+                CONF_THINKING_BUDGET: 0,
             },
         )
 
@@ -183,7 +187,11 @@ async def test_generate_structured_data_legacy_tools(
         hass.config_entries.async_update_subentry(
             mock_config_entry,
             subentry,
-            data={"chat_model": "claude-sonnet-4-0", "web_search": True},
+            data={
+                "chat_model": "claude-sonnet-4-0",
+                "web_search": True,
+                "thinking_budget": 0,
+            },
         )
 
     result = await ai_task.async_generate_data(

tests/components/anthropic/test_config_flow.py

@@ -330,7 +330,7 @@ async def test_subentry_web_search_user_location(
         "recommended": False,
         "region": "California",
         "temperature": 1.0,
-        "thinking_budget": 0,
+        "thinking_budget": 1024,
         "timezone": "America/Los_Angeles",
         "tool_search": False,
         "user_location": True,
@@ -487,7 +487,7 @@ async def test_model_list_error(
                 CONF_TEMPERATURE: 1.0,
                 CONF_CHAT_MODEL: "claude-haiku-4-5",
                 CONF_MAX_TOKENS: DEFAULT[CONF_MAX_TOKENS],
-                CONF_THINKING_BUDGET: 0,
+                CONF_THINKING_BUDGET: DEFAULT[CONF_THINKING_BUDGET],
                 CONF_WEB_SEARCH: False,
                 CONF_WEB_SEARCH_MAX_USES: 10,
                 CONF_WEB_SEARCH_USER_LOCATION: False,
@@ -614,7 +614,7 @@ async def test_model_list_error(
                 CONF_TEMPERATURE: 0.3,
                 CONF_CHAT_MODEL: DEFAULT[CONF_CHAT_MODEL],
                 CONF_MAX_TOKENS: DEFAULT[CONF_MAX_TOKENS],
-                CONF_THINKING_BUDGET: 0,
+                CONF_THINKING_BUDGET: DEFAULT[CONF_THINKING_BUDGET],
                 CONF_WEB_SEARCH: False,
                 CONF_WEB_SEARCH_MAX_USES: 5,
                 CONF_WEB_SEARCH_USER_LOCATION: False,
@@ -788,7 +788,7 @@ async def test_creating_ai_task_subentry_advanced(
         result["flow_id"],
         {
             CONF_CHAT_MODEL: "claude-sonnet-4-5",
-            CONF_MAX_TOKENS: 200,
+            CONF_MAX_TOKENS: 1200,
             CONF_TEMPERATURE: 0.5,
         },
     )
@@ -809,13 +809,13 @@ async def test_creating_ai_task_subentry_advanced(
     assert result4.get("data") == {
         CONF_RECOMMENDED: False,
         CONF_CHAT_MODEL: "claude-sonnet-4-5",
-        CONF_MAX_TOKENS: 200,
+        CONF_MAX_TOKENS: 1200,
         CONF_TEMPERATURE: 0.5,
         CONF_TOOL_SEARCH: False,
         CONF_WEB_SEARCH: False,
         CONF_WEB_SEARCH_MAX_USES: 5,
         CONF_WEB_SEARCH_USER_LOCATION: False,
-        CONF_THINKING_BUDGET: 0,
+        CONF_THINKING_BUDGET: 1024,
         CONF_CODE_EXECUTION: False,
         CONF_PROMPT_CACHING: "prompt",
     }

tests/components/anthropic/test_conversation.py

@@ -706,6 +706,21 @@ async def test_extended_thinking(
     assert call_args == snapshot
 
 
+@pytest.mark.parametrize(
+    "subentry_data",
+    [
+        {
+            CONF_LLM_HASS_API: "assist",
+            CONF_CHAT_MODEL: "claude-haiku-4-5",
+            CONF_THINKING_BUDGET: 0,
+        },
+        {
+            CONF_LLM_HASS_API: "assist",
+            CONF_CHAT_MODEL: "claude-opus-4-6",
+            CONF_THINKING_EFFORT: "none",
+        },
+    ],
+)
 @freeze_time("2024-05-24 12:00:00")
 async def test_disabled_thinking(
     hass: HomeAssistant,
@@ -713,16 +728,13 @@ async def test_disabled_thinking(
     mock_init_component,
     mock_create_stream: AsyncMock,
     snapshot: SnapshotAssertion,
+    subentry_data: dict[str, Any],
 ) -> None:
     """Test conversation with thinking effort disabled."""
     hass.config_entries.async_update_subentry(
         mock_config_entry,
         next(iter(mock_config_entry.subentries.values())),
-        data={
-            CONF_LLM_HASS_API: "assist",
-            CONF_CHAT_MODEL: "claude-opus-4-6",
-            CONF_THINKING_EFFORT: "none",
-        },
+        data=subentry_data,
     )
 
     mock_create_stream.return_value = [