iac-code 0.11.1__tar.gz → 0.11.2__tar.gz

{iac_code-0.11.1 → iac_code-0.11.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: iac_code
-Version: 0.11.1
+Version: 0.11.2
 Summary: Your AI-powered Infrastructure as Code assistant
 Classifier: Programming Language :: Python :: 3
 Classifier: Programming Language :: Python :: 3.10

iac_code-0.11.2/src/iac_code/__init__.py

@@ -0,0 +1,2 @@
+__version__ = "0.11.2"
+__release_date__ = "2026-08-03"

{iac_code-0.11.1 → iac_code-0.11.2}/src/iac_code/a2a/transports/dispatcher.py

@@ -795,10 +795,22 @@ class IacCodeRequestHandler(DefaultRequestHandler):
             raise TaskNotFoundError(f"Task {params.id} not found")
         if isinstance(self.task_store, A2ATaskStore) and not await self.task_store.is_task_active(params.id):
             raise TaskNotFoundError(f"Task {params.id} is not active")
+        terminal_state_seen = False
         async for event in super().on_subscribe_to_task(params, context):
+            event_state = _task_event_state(event)
+            terminal_state_seen = terminal_state_seen or event_state in TERMINAL_TASK_STATES
             yield event
-            if _task_event_state(event) in INTERRUPTED_TASK_STATES:
+            if event_state in INTERRUPTED_TASK_STATES:
                 return
+        if terminal_state_seen:
+            return
+
+        # a2a-sdk 1.1 can close its subscriber queue after the producer finishes but
+        # before the consumer publishes the last update. Recover the persisted terminal
+        # snapshot so subscribers do not observe a silent, non-terminal stream ending.
+        final_task = await self.task_store.get(params.id, context)
+        if final_task is not None and final_task.status.state in TERMINAL_TASK_STATES:
+            yield final_task
 
     async def on_create_task_push_notification_config(
         self, params: TaskPushNotificationConfig, context

{iac_code-0.11.1 → iac_code-0.11.2}/src/iac_code/config.py

@@ -19,7 +19,7 @@ from iac_code.i18n import _
 from iac_code.utils.file_security import atomic_write_text, ensure_private_dir, ensure_private_file
 
 # Default LLM model used when no model is saved in settings
-DEFAULT_MODEL = "qwen3.7-max"
+DEFAULT_MODEL = "qwen3.8-max"
 
 # Configuration directory
 _CONFIG_DIR_NAME = ".iac-code"

{iac_code-0.11.1 → iac_code-0.11.2}/src/iac_code/i18n/locales/de/LC_MESSAGES/messages.po

@@ -4,7 +4,7 @@
 #
 msgid ""
 msgstr ""
-"Project-Id-Version: iac-code 0.11.1\n"
+"Project-Id-Version: iac-code 0.11.2\n"
 "Report-Msgid-Bugs-To: \n"
 "POT-Creation-Date: 2026-06-04 15:58+0800\n"
 "PO-Revision-Date: 2026-05-13 00:00+0000\n"
@@ -7693,6 +7693,33 @@ msgstr "Fn::GetAtt verweist auf nicht vorhandene Ressource {}."
 msgid "The resource logical name or Count instance name is invalid."
 msgstr "Der logische Ressourcenname oder Count-Instanzname ist ungültig."
 
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+#, python-brace-format
+msgid ""
+"Fn::GetAtt for nested stack {} must use "
+"Outputs.<nested_stack_output_name>."
+msgstr ""
+"Fn::GetAtt für den verschachtelten Stack {} muss "
+"Outputs.<nested_stack_output_name> verwenden."
+
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+msgid ""
+"ALIYUN::ROS::Stack exposes child stack outputs only through the fixed "
+"Outputs. prefix, followed by a non-empty output name."
+msgstr ""
+"ALIYUN::ROS::Stack stellt Ausgaben des untergeordneten Stacks nur über "
+"das feste Präfix Outputs. bereit, gefolgt von einem nicht leeren "
+"Ausgabenamen."
+
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+#, python-brace-format
+msgid ""
+"Change the second argument to Outputs.<nested_stack_output_name>, for "
+"example Fn::GetAtt: [{}, Outputs.MyOutput]."
+msgstr ""
+"Ändern Sie das zweite Argument in Outputs.<nested_stack_output_name>, zum"
+" Beispiel Fn::GetAtt: [{}, Outputs.MyOutput]."
+
 #: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
 #, python-brace-format
 msgid ""

{iac_code-0.11.1 → iac_code-0.11.2}/src/iac_code/i18n/locales/es/LC_MESSAGES/messages.po

@@ -4,7 +4,7 @@
 #
 msgid ""
 msgstr ""
-"Project-Id-Version: iac-code 0.11.1\n"
+"Project-Id-Version: iac-code 0.11.2\n"
 "Report-Msgid-Bugs-To: \n"
 "POT-Creation-Date: 2026-06-04 15:58+0800\n"
 "PO-Revision-Date: 2026-05-13 00:00+0000\n"
@@ -7646,6 +7646,33 @@ msgstr "Fn::GetAtt referencias recurso inexistente {}."
 msgid "The resource logical name or Count instance name is invalid."
 msgstr "El nombre lógico de recurso o el nombre de instancia Count es inválido."
 
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+#, python-brace-format
+msgid ""
+"Fn::GetAtt for nested stack {} must use "
+"Outputs.<nested_stack_output_name>."
+msgstr ""
+"Fn::GetAtt para la pila anidada {} debe usar "
+"Outputs.<nested_stack_output_name>."
+
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+msgid ""
+"ALIYUN::ROS::Stack exposes child stack outputs only through the fixed "
+"Outputs. prefix, followed by a non-empty output name."
+msgstr ""
+"ALIYUN::ROS::Stack expone las salidas de la pila secundaria únicamente "
+"mediante el prefijo fijo Outputs., seguido de un nombre de salida no "
+"vacío."
+
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+#, python-brace-format
+msgid ""
+"Change the second argument to Outputs.<nested_stack_output_name>, for "
+"example Fn::GetAtt: [{}, Outputs.MyOutput]."
+msgstr ""
+"Cambie el segundo argumento a Outputs.<nested_stack_output_name>, por "
+"ejemplo Fn::GetAtt: [{}, Outputs.MyOutput]."
+
 #: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
 #, python-brace-format
 msgid ""

{iac_code-0.11.1 → iac_code-0.11.2}/src/iac_code/i18n/locales/fr/LC_MESSAGES/messages.po

@@ -4,7 +4,7 @@
 #
 msgid ""
 msgstr ""
-"Project-Id-Version: iac-code 0.11.1\n"
+"Project-Id-Version: iac-code 0.11.2\n"
 "Report-Msgid-Bugs-To: \n"
 "POT-Creation-Date: 2026-06-04 15:58+0800\n"
 "PO-Revision-Date: 2026-05-13 00:00+0000\n"
@@ -7684,6 +7684,32 @@ msgstr "Fn::GetAtt références ressource inexistante {}."
 msgid "The resource logical name or Count instance name is invalid."
 msgstr "Le nom logique de la ressource ou le nom d'instance Count est invalide."
 
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+#, python-brace-format
+msgid ""
+"Fn::GetAtt for nested stack {} must use "
+"Outputs.<nested_stack_output_name>."
+msgstr ""
+"Fn::GetAtt pour la pile imbriquée {} doit utiliser "
+"Outputs.<nested_stack_output_name>."
+
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+msgid ""
+"ALIYUN::ROS::Stack exposes child stack outputs only through the fixed "
+"Outputs. prefix, followed by a non-empty output name."
+msgstr ""
+"ALIYUN::ROS::Stack expose les sorties de la pile enfant uniquement via le"
+" préfixe fixe Outputs., suivi d’un nom de sortie non vide."
+
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+#, python-brace-format
+msgid ""
+"Change the second argument to Outputs.<nested_stack_output_name>, for "
+"example Fn::GetAtt: [{}, Outputs.MyOutput]."
+msgstr ""
+"Remplacez le deuxième argument par Outputs.<nested_stack_output_name>, "
+"par exemple Fn::GetAtt: [{}, Outputs.MyOutput]."
+
 #: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
 #, python-brace-format
 msgid ""

{iac_code-0.11.1 → iac_code-0.11.2}/src/iac_code/i18n/locales/ja/LC_MESSAGES/messages.po

@@ -4,7 +4,7 @@
 #
 msgid ""
 msgstr ""
-"Project-Id-Version: iac-code 0.11.1\n"
+"Project-Id-Version: iac-code 0.11.2\n"
 "Report-Msgid-Bugs-To: \n"
 "POT-Creation-Date: 2026-06-04 15:58+0800\n"
 "PO-Revision-Date: 2026-05-13 00:00+0000\n"
@@ -7156,6 +7156,32 @@ msgstr "Fn::GetAtt は、存在しないリソース {} を参照します。"
 msgid "The resource logical name or Count instance name is invalid."
 msgstr "リソース論理名またはCountインスタンス名は無効です。"
 
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+#, python-brace-format
+msgid ""
+"Fn::GetAtt for nested stack {} must use "
+"Outputs.<nested_stack_output_name>."
+msgstr ""
+"ネストされたスタック {} の Fn::GetAtt では Outputs.<nested_stack_output_name> "
+"を使用する必要があります。"
+
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+msgid ""
+"ALIYUN::ROS::Stack exposes child stack outputs only through the fixed "
+"Outputs. prefix, followed by a non-empty output name."
+msgstr ""
+"ALIYUN::ROS::Stack では、固定プレフィックス Outputs. "
+"の後に空でない出力名を指定した場合にのみ、子スタックの出力を参照できます。"
+
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+#, python-brace-format
+msgid ""
+"Change the second argument to Outputs.<nested_stack_output_name>, for "
+"example Fn::GetAtt: [{}, Outputs.MyOutput]."
+msgstr ""
+"2 番目の引数を Outputs.<nested_stack_output_name> に変更してください。例: Fn::GetAtt: [{},"
+" Outputs.MyOutput]。"
+
 #: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
 #, python-brace-format
 msgid ""

{iac_code-0.11.1 → iac_code-0.11.2}/src/iac_code/i18n/locales/pt/LC_MESSAGES/messages.po

@@ -4,7 +4,7 @@
 #
 msgid ""
 msgstr ""
-"Project-Id-Version: iac-code 0.11.1\n"
+"Project-Id-Version: iac-code 0.11.2\n"
 "Report-Msgid-Bugs-To: \n"
 "POT-Creation-Date: 2026-06-04 15:58+0800\n"
 "PO-Revision-Date: 2026-05-13 00:00+0000\n"
@@ -7571,6 +7571,32 @@ msgstr "Referências Fn::GetAtt recurso inexistente {}."
 msgid "The resource logical name or Count instance name is invalid."
 msgstr "O nome lógico do recurso ou o nome da instância Count é inválido."
 
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+#, python-brace-format
+msgid ""
+"Fn::GetAtt for nested stack {} must use "
+"Outputs.<nested_stack_output_name>."
+msgstr ""
+"Fn::GetAtt para a pilha aninhada {} deve usar "
+"Outputs.<nested_stack_output_name>."
+
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+msgid ""
+"ALIYUN::ROS::Stack exposes child stack outputs only through the fixed "
+"Outputs. prefix, followed by a non-empty output name."
+msgstr ""
+"ALIYUN::ROS::Stack expõe as saídas da pilha filha somente por meio do "
+"prefixo fixo Outputs., seguido de um nome de saída não vazio."
+
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+#, python-brace-format
+msgid ""
+"Change the second argument to Outputs.<nested_stack_output_name>, for "
+"example Fn::GetAtt: [{}, Outputs.MyOutput]."
+msgstr ""
+"Altere o segundo argumento para Outputs.<nested_stack_output_name>, por "
+"exemplo, Fn::GetAtt: [{}, Outputs.MyOutput]."
+
 #: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
 #, python-brace-format
 msgid ""

{iac_code-0.11.1 → iac_code-0.11.2}/src/iac_code/i18n/locales/zh/LC_MESSAGES/messages.po

@@ -4,7 +4,7 @@
 #
 msgid ""
 msgstr ""
-"Project-Id-Version: iac-code 0.11.1\n"
+"Project-Id-Version: iac-code 0.11.2\n"
 "Report-Msgid-Bugs-To: \n"
 "POT-Creation-Date: 2026-06-04 15:58+0800\n"
 "PO-Revision-Date: 2026-04-02 00:00+0000\n"
@@ -6993,6 +6993,28 @@ msgstr "Fn::GetAtt 引用了不存在的资源 {}。"
 msgid "The resource logical name or Count instance name is invalid."
 msgstr "资源逻辑名或 Count 实例名无效。"
 
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+#, python-brace-format
+msgid ""
+"Fn::GetAtt for nested stack {} must use "
+"Outputs.<nested_stack_output_name>."
+msgstr "嵌套资源栈 {} 的 Fn::GetAtt 必须使用 Outputs.<nested_stack_output_name>。"
+
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+msgid ""
+"ALIYUN::ROS::Stack exposes child stack outputs only through the fixed "
+"Outputs. prefix, followed by a non-empty output name."
+msgstr "ALIYUN::ROS::Stack 只能通过固定的 Outputs. 前缀访问子资源栈输出，前缀后必须跟一个非空的输出名称。"
+
+#: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
+#, python-brace-format
+msgid ""
+"Change the second argument to Outputs.<nested_stack_output_name>, for "
+"example Fn::GetAtt: [{}, Outputs.MyOutput]."
+msgstr ""
+"请将第二个参数改为 Outputs.<nested_stack_output_name>，例如 Fn::GetAtt: [{}, "
+"Outputs.MyOutput]。"
+
 #: src/iac_code/tools/cloud/aliyun/ros_validation/analyzer.py
 #, python-brace-format
 msgid ""

{iac_code-0.11.1 → iac_code-0.11.2}/src/iac_code/pipeline/engine/handoff.py

@@ -43,5 +43,18 @@ def build_handoff_summary(
     if missing:
         lines.extend(["", "Missing context fields:"])
         lines.extend(f"- {field_name}" for field_name in missing)
-    lines.extend(["", "Use this context when answering follow-up questions after the pipeline handoff."])
+    lines.extend(
+        [
+            "",
+            "Safety requirements for normal chat:",
+            (
+                "- Before performing any operation that releases, deletes, or otherwise destroys a resource, "
+                "obtain a fresh, explicit confirmation from the user in normal chat. Any confirmation given "
+                "during the pipeline does not count."
+            ),
+            ("- Exception: pipeline-managed automatic cleanup may proceed without this additional confirmation."),
+            "",
+            "Use this context when answering follow-up questions after the pipeline handoff.",
+        ]
+    )
     return "\n".join(lines)

{iac_code-0.11.1 → iac_code-0.11.2}/src/iac_code/providers/anthropic_provider.py

@@ -144,7 +144,12 @@ class AnthropicProvider(Provider):
 
         kwargs: dict[str, Any] = {}
         if self._thinking_disabled():
-            if not spec.adaptive_always_on:
+            forbidden_disable_efforts = {item.value for item in spec.disable_forbidden_efforts}
+            if effort in forbidden_disable_efforts:
+                raise ValueError(
+                    f"Anthropic thinking cannot be disabled for {self._model} at {effort} effort; use high or lower"
+                )
+            if spec.supports_disable:
                 kwargs["thinking"] = {"type": "disabled"}
         elif not spec.adaptive_always_on and (effort is not None or self._thinking_forced()):
             kwargs["thinking"] = {"type": "adaptive"}

{iac_code-0.11.1 → iac_code-0.11.2}/src/iac_code/providers/dashscope_provider.py

@@ -17,7 +17,7 @@ DASHSCOPE_TOKEN_PLAN_BASE_URL = "https://token-plan.cn-beijing.maas.aliyuncs.com
 # Prefix-matched against the model name.  Extend when new models are added.
 # Ref: https://help.aliyun.com/zh/model-studio/context-cache
 _EXPLICIT_CACHE_MODEL_PREFIXES: tuple[str, ...] = (
-    "qwen3.8-max-preview",
+    "qwen3.8-max",
     "qwen3.7-max",
     "qwen3.7-plus",
     "qwen3-coder-plus",
@@ -33,7 +33,7 @@ _EXPLICIT_CACHE_MODEL_PREFIXES: tuple[str, ...] = (
 # Models documented to accept extra_body.preserve_thinking. Keep this list
 # separate from thinking support: sending the parameter to other models fails.
 _PRESERVE_THINKING_MODEL_PREFIXES: tuple[str, ...] = (
-    "qwen3.8-max-preview",
+    "qwen3.8-max",
     "qwen3.7-max",
     "qwen3.7-plus",
     "qwen3.6-max-preview",

{iac_code-0.11.1 → iac_code-0.11.2}/src/iac_code/providers/deepseek_provider.py

@@ -14,7 +14,9 @@ class DeepSeekProvider(OpenAIProvider):
 
     DeepSeek uses ``reasoning_effort`` plus
     ``extra_body.thinking.type=enabled``; the registry's ``allowed_efforts``
-    constrains the effort vocabulary to ``high`` / ``max`` for DeepSeek V4.
+    constrains the effort vocabulary to ``low`` / ``high`` / ``max`` for
+    DeepSeek V4. Non-thinking mode is selected with
+    ``extra_body.thinking.type=disabled``.
 
     Reasoning content is captured via ``reasoning_content`` in the stream
     and echoed back as ``reasoning_content`` on subsequent assistant
@@ -48,6 +50,8 @@ class DeepSeekProvider(OpenAIProvider):
         )
 
     def _build_thinking_kwargs(self) -> dict[str, Any]:
+        if self._thinking_disabled():
+            return {"extra_body": {"thinking": {"type": "disabled"}}}
         kwargs = super()._build_thinking_kwargs()
         if not kwargs:
             return {}

{iac_code-0.11.1 → iac_code-0.11.2}/src/iac_code/providers/manager.py

@@ -304,6 +304,7 @@ def _error_event_from_exception(exc: BaseException) -> ErrorEvent:
 
 MODEL_FALLBACK_MAP = {
     "claude-fable-5": "claude-opus-4-8",
+    "claude-opus-5": "claude-opus-4-8",
     "claude-opus-4-8": "claude-sonnet-5",
     "claude-opus-4-7": "claude-haiku-4-5-20251001",
     "claude-opus-4-6": "claude-haiku-4-5-20251001",
@@ -315,17 +316,20 @@ MODEL_FALLBACK_MAP = {
     "gpt-5.6-terra": "gpt-5.6-luna",
     "gpt-5.5": "gpt-5.4",
     "gpt-5.4": "gpt-5.4-mini",
-    "qwen3.8-max-preview": "qwen3.7-plus",
+    "qwen3.8-max": "qwen3.7-plus",
+    "qwen3.8-max-preview": "qwen3.8-max",
     "qwen3.7-max": "qwen3.7-plus",
     "kimi/kimi-k3": "kimi-k2.7-code",
     "kimi-k3": "kimi-k2.7-code",
     "glm-5.2-fast-preview": "glm-5.2",
     "glm-5.2": "glm-5.1",
     "deepseek-v4-pro": "deepseek-v4-flash",
+    "deepseek-v4-flash-0731": "deepseek-v4-flash",
 }
 
 _MODEL_REFUSAL_FALLBACK_MAP = {
     "claude-fable-5": "claude-opus-4-8",
+    "claude-opus-5": "claude-opus-4-8",
 }
 
 _PROVIDER_MODEL_FALLBACK_MAP = {

{iac_code-0.11.1 → iac_code-0.11.2}/src/iac_code/providers/registry.py

@@ -45,7 +45,8 @@ PROVIDER_REGISTRY: dict[str, ProviderDescriptor] = {
         provider_class="iac_code.providers.dashscope_provider.DashScopeProvider",
         base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
         models=[
-            ModelEntry("qwen3.7-max", is_default=True),
+            ModelEntry("qwen3.8-max", is_default=True, support_multimodal=True),
+            ModelEntry("qwen3.7-max"),
             ModelEntry("qwen3.7-plus", support_multimodal=True),
             ModelEntry("qwen3.6-plus", support_multimodal=True),
             ModelEntry("qwen3.6-flash", support_multimodal=True),
@@ -65,6 +66,7 @@ PROVIDER_REGISTRY: dict[str, ProviderDescriptor] = {
             ModelEntry("kimi-k2.6", support_multimodal=True),
             ModelEntry("kimi-k2.5", support_multimodal=True),
             ModelEntry("deepseek-v4-pro"),
+            ModelEntry("deepseek-v4-flash-0731"),
             ModelEntry("deepseek-v4-flash"),
             ModelEntry("glm-5.2-fast-preview"),
             ModelEntry("glm-5.2"),
@@ -81,12 +83,14 @@ PROVIDER_REGISTRY: dict[str, ProviderDescriptor] = {
         provider_class="iac_code.providers.dashscope_provider.DashScopeProvider",
         base_url="https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1",
         models=[
-            ModelEntry("qwen3.8-max-preview", is_default=True, support_multimodal=True),
+            ModelEntry("qwen3.8-max", is_default=True, support_multimodal=True),
+            ModelEntry("qwen3.8-max-preview", support_multimodal=True),
             ModelEntry("qwen3.7-max"),
             ModelEntry("qwen3.7-plus", support_multimodal=True),
             ModelEntry("qwen3.6-plus", support_multimodal=True),
             ModelEntry("qwen3.6-flash", support_multimodal=True),
             ModelEntry("deepseek-v4-pro"),
+            ModelEntry("deepseek-v4-flash-0731"),
             ModelEntry("deepseek-v4-flash"),
             ModelEntry("deepseek-v3.2"),
             ModelEntry("glm-5.2"),
@@ -129,6 +133,7 @@ PROVIDER_REGISTRY: dict[str, ProviderDescriptor] = {
         base_url=None,
         models=[
             ModelEntry("claude-fable-5", is_default=True, support_multimodal=True),
+            ModelEntry("claude-opus-5", support_multimodal=True),
             ModelEntry("claude-opus-4-8", support_multimodal=True),
             ModelEntry("claude-sonnet-5", support_multimodal=True),
             ModelEntry("claude-opus-4-7", support_multimodal=True),

{iac_code-0.11.1 → iac_code-0.11.2}/src/iac_code/providers/thinking.py

@@ -77,6 +77,8 @@ class ThinkingSpec:
     uses_reasoning_effort_param: bool = False
     adaptive_always_on: bool = False
     supports_disable: bool = True
+    thinking_enabled_by_default: bool = False
+    disable_forbidden_efforts: tuple[EffortLevel, ...] = ()
 
     @property
     def supports_effort(self) -> bool:
@@ -155,8 +157,24 @@ _ANTHROPIC_46_EFFORTS: tuple[EffortLevel, ...] = (
     EffortLevel.AUTO,
 )
 
-# DeepSeek V4 accepts only high/max — XHIGH is intentionally skipped.
-_DEEPSEEK_EFFORTS: tuple[EffortLevel, ...] = (EffortLevel.HIGH, EffortLevel.MAX)
+# The official DeepSeek endpoint exposes low/high/max — MEDIUM and XHIGH are
+# intentionally skipped because the API rejects them.
+_DEEPSEEK_EFFORTS: tuple[EffortLevel, ...] = (
+    EffortLevel.LOW,
+    EffortLevel.HIGH,
+    EffortLevel.MAX,
+)
+
+# DashScope accepts the full effort vocabulary for its hosted DeepSeek V4
+# models. Low/medium currently map to high and xhigh maps to max server-side,
+# but keeping the accepted wire values lets callers migrate without a 400.
+_DASHSCOPE_DEEPSEEK_EFFORTS: tuple[EffortLevel, ...] = (
+    EffortLevel.LOW,
+    EffortLevel.MEDIUM,
+    EffortLevel.HIGH,
+    EffortLevel.XHIGH,
+    EffortLevel.MAX,
+)
 
 _GLM_EFFORTS: tuple[EffortLevel, ...] = (
     EffortLevel.NONE,
@@ -209,11 +227,20 @@ _DASHSCOPE_GLM51_SPEC = ThinkingSpec(
 )
 
 _DASHSCOPE_QWEN38_SPEC = ThinkingSpec(
+    ThinkingFamily.DASHSCOPE,
+    (EffortLevel.LOW, EffortLevel.MEDIUM, EffortLevel.XHIGH),
+    EffortLevel.XHIGH,
+    uses_reasoning_effort_param=True,
+    thinking_enabled_by_default=True,
+)
+
+_DASHSCOPE_QWEN38_PREVIEW_SPEC = ThinkingSpec(
     ThinkingFamily.DASHSCOPE,
     (EffortLevel.LOW, EffortLevel.MEDIUM, EffortLevel.XHIGH),
     EffortLevel.XHIGH,
     uses_reasoning_effort_param=True,
     supports_disable=False,
+    thinking_enabled_by_default=True,
 )
 
 _DASHSCOPE_KIMI_K3_SPEC = ThinkingSpec(
@@ -230,6 +257,14 @@ _ANTHROPIC_ADAPTIVE_SPEC = ThinkingSpec(
     EffortLevel.HIGH,
 )
 
+_ANTHROPIC_OPUS5_SPEC = ThinkingSpec(
+    ThinkingFamily.ANTHROPIC_ADAPTIVE,
+    _ANTHROPIC_MODERN_EFFORTS,
+    EffortLevel.HIGH,
+    thinking_enabled_by_default=True,
+    disable_forbidden_efforts=(EffortLevel.XHIGH, EffortLevel.MAX),
+)
+
 _ANTHROPIC_46_ADAPTIVE_SPEC = ThinkingSpec(
     ThinkingFamily.ANTHROPIC_ADAPTIVE,
     _ANTHROPIC_46_EFFORTS,
@@ -250,6 +285,12 @@ _OPENAI_GPT55_SPEC = ThinkingSpec(ThinkingFamily.OPENAI, _OPENAI_EFFORTS, Effort
 _OPENAI_REASONING_SPEC = ThinkingSpec(ThinkingFamily.OPENAI, _OPENAI_EFFORTS, EffortLevel.HIGH)
 _OPENAI_CODEX_SPEC = ThinkingSpec(ThinkingFamily.OPENAI, _OPENAI_CODEX_EFFORTS, EffortLevel.MEDIUM)
 _OPENAI_O_SERIES_SPEC = ThinkingSpec(ThinkingFamily.OPENAI, _OPENAI_O_SERIES_EFFORTS, EffortLevel.HIGH)
+_DEEPSEEK_SPEC = ThinkingSpec(
+    ThinkingFamily.OPENAI,
+    _DEEPSEEK_EFFORTS,
+    EffortLevel.HIGH,
+    thinking_enabled_by_default=True,
+)
 _KIMI_K3_SPEC = ThinkingSpec(ThinkingFamily.KIMI, _KIMI_K3_EFFORTS, EffortLevel.MAX)
 _ZHIPU_GLM52_SPEC = ThinkingSpec(
     ThinkingFamily.ZHIPU,
@@ -262,6 +303,7 @@ _ZHIPU_GLM52_SPEC = ThinkingSpec(
 MODEL_THINKING: dict[str, dict[str, ThinkingSpec]] = {
     "anthropic": {
         "claude-fable-5": _ANTHROPIC_ADAPTIVE_ALWAYS_ON_SPEC,
+        "claude-opus-5": _ANTHROPIC_OPUS5_SPEC,
         "claude-sonnet-5": _ANTHROPIC_ADAPTIVE_SPEC,
         "claude-opus-4-8": _ANTHROPIC_ADAPTIVE_SPEC,
         "claude-opus-4-7": _ANTHROPIC_ADAPTIVE_SPEC,
@@ -291,10 +333,11 @@ MODEL_THINKING: dict[str, dict[str, ThinkingSpec]] = {
         "o4-mini": _OPENAI_O_SERIES_SPEC,
     },
     "deepseek": {
-        "deepseek-v4-pro": ThinkingSpec(ThinkingFamily.OPENAI, _DEEPSEEK_EFFORTS, EffortLevel.HIGH),
-        "deepseek-v4-flash": ThinkingSpec(ThinkingFamily.OPENAI, _DEEPSEEK_EFFORTS, EffortLevel.HIGH),
+        "deepseek-v4-pro": _DEEPSEEK_SPEC,
+        "deepseek-v4-flash": _DEEPSEEK_SPEC,
     },
     "dashscope": {
+        "qwen3.8-max": _DASHSCOPE_QWEN38_SPEC,
         "qwen3.7-max": ThinkingSpec(ThinkingFamily.DASHSCOPE),
         "qwen3.7-plus": ThinkingSpec(ThinkingFamily.DASHSCOPE),
         "qwen3.6-max-preview": ThinkingSpec(ThinkingFamily.DASHSCOPE),
@@ -314,16 +357,34 @@ MODEL_THINKING: dict[str, dict[str, ThinkingSpec]] = {
         "glm-5.1": _DASHSCOPE_GLM51_SPEC,
         "MiniMax-M2.5": ThinkingSpec(ThinkingFamily.DASHSCOPE),
         "MiniMax/MiniMax-M3": ThinkingSpec(ThinkingFamily.MINIMAX),
-        "deepseek-v4-pro": ThinkingSpec(ThinkingFamily.DASHSCOPE, _DEEPSEEK_EFFORTS, EffortLevel.HIGH),
-        "deepseek-v4-flash": ThinkingSpec(ThinkingFamily.DASHSCOPE, _DEEPSEEK_EFFORTS, EffortLevel.HIGH),
+        "deepseek-v4-pro": ThinkingSpec(
+            ThinkingFamily.DASHSCOPE,
+            _DASHSCOPE_DEEPSEEK_EFFORTS,
+            EffortLevel.HIGH,
+            uses_reasoning_effort_param=True,
+        ),
+        "deepseek-v4-flash-0731": ThinkingSpec(
+            ThinkingFamily.DASHSCOPE,
+            _DASHSCOPE_DEEPSEEK_EFFORTS,
+            EffortLevel.HIGH,
+            uses_reasoning_effort_param=True,
+        ),
+        "deepseek-v4-flash": ThinkingSpec(
+            ThinkingFamily.DASHSCOPE,
+            _DASHSCOPE_DEEPSEEK_EFFORTS,
+            EffortLevel.HIGH,
+            uses_reasoning_effort_param=True,
+        ),
     },
     "dashscope_token_plan": {
-        "qwen3.8-max-preview": _DASHSCOPE_QWEN38_SPEC,
+        "qwen3.8-max": _DASHSCOPE_QWEN38_SPEC,
+        "qwen3.8-max-preview": _DASHSCOPE_QWEN38_PREVIEW_SPEC,
         "qwen3.7-max": ThinkingSpec(ThinkingFamily.DASHSCOPE),
         "qwen3.7-plus": ThinkingSpec(ThinkingFamily.DASHSCOPE),
         "qwen3.6-plus": ThinkingSpec(ThinkingFamily.DASHSCOPE),
         "qwen3.6-flash": ThinkingSpec(ThinkingFamily.DASHSCOPE),
         "deepseek-v4-pro": ThinkingSpec(ThinkingFamily.DASHSCOPE, _DEEPSEEK_EFFORTS, EffortLevel.HIGH),
+        "deepseek-v4-flash-0731": ThinkingSpec(ThinkingFamily.DASHSCOPE, _DEEPSEEK_EFFORTS, EffortLevel.HIGH),
         "deepseek-v4-flash": ThinkingSpec(ThinkingFamily.DASHSCOPE, _DEEPSEEK_EFFORTS, EffortLevel.HIGH),
         "deepseek-v3.2": ThinkingSpec(ThinkingFamily.DASHSCOPE),
         "glm-5.1": _DASHSCOPE_GLM51_SPEC,
@@ -473,9 +534,9 @@ _THINKING_DISABLE_EFFORTS: frozenset[str] = frozenset({"none", "off", "disable",
 # Families whose providers emit an explicit "thinking on" directive by default
 # when the user has NOT configured ``thinkingEnabled`` (DashScope
 # ``enable_thinking=True``, Kimi/Zhipu ``thinking.type=enabled``). For these an
-# unset config still means the next turn thinks. Reasoning-effort families
-# (OpenAI/Anthropic/Gemini) instead emit nothing when unset and let the server
-# decide — unobservable here, so they resolve to off.
+# unset config still means the next turn thinks. Most reasoning-effort models
+# (OpenAI/Anthropic/Gemini) instead emit nothing when unset and resolve to off;
+# models with a documented server-side thinking default opt in explicitly.
 _DEFAULT_ON_FAMILIES: frozenset[ThinkingFamily] = frozenset(
     {ThinkingFamily.DASHSCOPE, ThinkingFamily.KIMI, ThinkingFamily.ZHIPU}
 )
@@ -498,15 +559,19 @@ def resolve_thinking_active(
     spec = get_thinking_spec(provider_key or "", model or "")
     if spec.family is ThinkingFamily.NONE:
         return False
+    normalized_effort = normalize_effort(effort)
     if thinking_enabled is False:
-        return False
+        if not spec.supports_disable:
+            return True
+        forbidden = {item.value for item in spec.disable_forbidden_efforts}
+        return normalized_effort in forbidden
     # DashScope disable-effort tokens force thinking off even when configured on.
-    if spec.family is ThinkingFamily.DASHSCOPE and normalize_effort(effort) in _THINKING_DISABLE_EFFORTS:
+    if spec.family is ThinkingFamily.DASHSCOPE and normalized_effort in _THINKING_DISABLE_EFFORTS:
         return False
     if thinking_enabled is True:
         return True
     # thinking_enabled is None (not configured): use the family default.
-    return spec.family in _DEFAULT_ON_FAMILIES
+    return spec.thinking_enabled_by_default or spec.adaptive_always_on or spec.family in _DEFAULT_ON_FAMILIES
 
 
 # Anthropic extended-thinking budget tokens per effort level.