core/agent+anthropic: drop the thinking block a model refuses to disable

src/plyngent/agent/messages_bridge.py

@@ -457,14 +457,15 @@ def _adaptive_thinking_from_param(
     return AnthropicThinkingConfig(type="adaptive"), AnthropicOutputConfig(effort=level)
 
 
-def extended_thinking_rejected(message: str) -> bool:
+def thinking_config_refused(message: str) -> bool:
-    """Whether a ``/messages`` failure is the API refusing extended thinking.
+    """Whether a ``/messages`` failure is the API refusing a request's thinking block.
 
-    The documented wording is ``"thinking.type.enabled" is not supported for this
+    The documented wordings are ``"thinking.type.enabled" is not supported for
-    model. Use "thinking.type.adaptive" and "output_config.effort" …``; gateways
+    this model. Use "thinking.type.adaptive" and "output_config.effort" …`` and
-    paraphrase it (``claude-opus-5-5 requires adaptive thinking; omit thinking or
+    the same sentence for ``"thinking.type.disabled"``; gateways paraphrase them
-    use thinking.type=adaptive …``), so match the one word they all carry —
+    (``claude-opus-5-5 requires adaptive thinking; omit thinking or use
-    callers only get here after a request that sent extended thinking.
+    thinking.type=adaptive …``), so match the one word they all carry — callers
+    only get here after a request that sent a thinking block.
     """
     return "adaptive" in message.lower()
 

src/plyngent/agent/messages_dispatch.py

@@ -11,6 +11,7 @@ from __future__ import annotations
 import json
 from typing import TYPE_CHECKING, Any, Final, cast
 
+import msgspec
 from msgspec import UNSET
 
 from plyngent.agent.messages_bridge import (
@@ -18,7 +19,7 @@ from plyngent.agent.messages_bridge import (
     anthropic_stop_to_finish_reason,
     anthropic_usage_to_dict,
     chat_param_to_anthropic_param,
-    extended_thinking_rejected,
+    thinking_config_refused,
 )
 from plyngent.agent.stream_chunks import (
     finish_reason_chunk,
@@ -56,46 +57,59 @@ def _thinking_budget(client: AnthropicClient) -> int:
     return max(0, int(getattr(client, "thinking_budget_tokens", 0)))
 
 
+# The thinking types a refusal has a fallback for: extended thinking becomes
+# adaptive, and a block disabling thinking is dropped (such a model thinks
+# whatever the request says).
+_RETRYABLE_THINKING_TYPES: Final = frozenset({"enabled", "disabled"})
+
+
 def _create_param(
     client: AnthropicClient,
     param: ChatCompletionsParam,
     provider_tools: Sequence[dict[str, Any]] | None,
-    *,
-    adaptive: bool | None = None,
 ) -> AnthropicMessagesParam:
-    """Build the ``/messages`` request for *param*.
+    """Build the ``/messages`` request for *param*, the way its model now takes it.
 
-    *adaptive* picks the thinking shape; ``None`` asks the client whether this
+    A model this session found refusing extended thinking gets adaptive thinking;
-    session already found *param*'s model rejecting extended thinking.
+    one that refuses being told to stop thinking is sent no block at all (its
+    thinking is always on, so the API answers a block that disables it with a
+    400 — see :func:`_retry_with_thinking_fallback`).
     """
-    use_adaptive = client.needs_adaptive_thinking(param.model) if adaptive is None else adaptive
+    model = param.model
-    return chat_param_to_anthropic_param(
+    create = chat_param_to_anthropic_param(
         param,
         thinking_budget_tokens=_thinking_budget(client),
         provider_tools=provider_tools,
-        adaptive_thinking=use_adaptive,
+        adaptive_thinking=client.refuses_thinking_type(model, "enabled"),
     )
+    if (
+        create.thinking is not UNSET
+        and create.thinking.type == "disabled"
+        and client.refuses_thinking_type(model, "disabled")
+    ):
+        return msgspec.structs.replace(create, thinking=UNSET)
+    return create
 
 
-def _retry_with_adaptive_thinking(
+def _retry_with_thinking_fallback(
     client: AnthropicClient,
     param: ChatCompletionsParam,
     provider_tools: Sequence[dict[str, Any]] | None,
     error: Exception,
 ) -> AnthropicMessagesParam | None:
-    """The request to retry with when the model refused extended thinking.
+    """The request to retry with when the model refused the thinking block.
 
-    ``None`` when the failure is not that refusal, or when the request never sent
+    ``None`` when the failure is not that refusal, or when there is nothing left
-    extended thinking. The model is remembered on the client, so only its first
+    to try. The refusal is remembered on the client, so only the model's first
     request pays the extra round trip.
     """
-    if not extended_thinking_rejected(str(error)):
+    if not thinking_config_refused(str(error)):
         return None
     create = _create_param(client, param, provider_tools)
-    if create.thinking is UNSET or create.thinking.type != "enabled":
+    if create.thinking is UNSET or create.thinking.type not in _RETRYABLE_THINKING_TYPES:
         return None
-    client.note_adaptive_thinking(param.model)
+    client.note_refused_thinking_type(param.model, create.thinking.type)
-    return _create_param(client, param, provider_tools, adaptive=True)
+    return _create_param(client, param, provider_tools)
 
 
 class _ToolBlockState:
@@ -416,7 +430,7 @@ async def _stream_events(
     *,
     provider_tools: Sequence[dict[str, Any]] | None = None,
 ) -> AsyncIterator[AnthropicStreamEvent]:
-    """Open the Messages stream, retrying once with adaptive thinking if refused.
+    """Open the Messages stream, retrying once with the shape the model takes.
 
     A refusal arrives before the stream yields anything, so the retry cannot
     duplicate events the caller has already seen.
@@ -430,7 +444,7 @@ async def _stream_events(
     except RuntimeError as exc:
         if saw_event:
             raise
-        retry = _retry_with_adaptive_thinking(client, param, provider_tools, exc)
+        retry = _retry_with_thinking_fallback(client, param, provider_tools, exc)
         if retry is None:
             raise
         async for event in await client.messages(retry, stream=True):
@@ -491,7 +505,7 @@ async def dispatch_messages(
     try:
         response = await client.messages(create, stream=False)
     except RuntimeError as exc:
-        retry = _retry_with_adaptive_thinking(client, param, provider_tools, exc)
+        retry = _retry_with_thinking_fallback(client, param, provider_tools, exc)
         if retry is None:
             raise
         response = await client.messages(retry, stream=False)

src/plyngent/lmproto/anthropic/client.py

@@ -36,9 +36,10 @@ class AnthropicClient:
     models_decoder: msgspec.json.Decoder[AnthropicModelsResponse]
     # Explicit ``thinking`` budget for this client (0 = derive from the effort).
     thinking_budget_tokens: int = 0
-    # Models this session found to reject extended thinking; the dispatch retries
+    # Per model id, the ``thinking.type`` values the API refused this session; the
-    # their requests with adaptive thinking instead (see ``messages_dispatch``).
+    # dispatch retries with the shape the model takes instead (see
-    adaptive_thinking_models: set[str]
+    # ``agent.messages_dispatch``).
+    refused_thinking_types: dict[str, set[str]]
     _api_key: str
     _api_version: str
 
@@ -49,17 +50,17 @@ class AnthropicClient:
         self.stream_decoder = msgspec.json.Decoder(AnthropicStreamEvent)
         self.models_decoder = msgspec.json.Decoder(AnthropicModelsResponse)
         self.thinking_budget_tokens = max(0, config.thinking_budget_tokens)
-        self.adaptive_thinking_models = set()
+        self.refused_thinking_types = {}
         self._api_key = config.api_key
         self._api_version = config.anthropic_version
 
-    def needs_adaptive_thinking(self, model: str) -> bool:
+    def refuses_thinking_type(self, model: str, thinking_type: str) -> bool:
-        """Whether *model* already rejected extended thinking in this session."""
+        """Whether the API already refused *thinking_type* for *model* this session."""
-        return model in self.adaptive_thinking_models
+        return thinking_type in self.refused_thinking_types.get(model, ())
 
-    def note_adaptive_thinking(self, model: str) -> None:
+    def note_refused_thinking_type(self, model: str, thinking_type: str) -> None:
-        """Remember that *model* requires adaptive thinking, saving the retry."""
+        """Remember that the API refuses *thinking_type* for *model*, saving the retry."""
-        self.adaptive_thinking_models.add(model)
+        self.refused_thinking_types.setdefault(model, set()).add(thinking_type)
 
     def _headers(self) -> dict[str, str]:
         return {

tests/test_agent/test_messages_bridge.py

@@ -14,8 +14,8 @@ from plyngent.agent.messages_bridge import (
     anthropic_usage_to_dict,
     chat_messages_to_anthropic,
     chat_param_to_anthropic_param,
-    extended_thinking_rejected,
     provider_tools_to_anthropic_tools,
+    thinking_config_refused,
     tool_items_to_anthropic_tools,
 )
 from plyngent.lmproto.anthropic.model import (
@@ -432,15 +432,20 @@ def test_chat_param_to_anthropic_param_adaptive_without_effort_sends_nothing() -
     assert body.output_config is UNSET
 
 
-def test_extended_thinking_rejected_matches_the_api_and_gateway_wording() -> None:
+def test_thinking_config_refused_matches_the_api_and_gateway_wording() -> None:
-    documented = (
+    documented_enabled = (
         'messages HTTP 400: {"error":{"message":"\\"thinking.type.enabled\\" is not supported for this '
         'model. Use \\"thinking.type.adaptive\\" and \\"output_config.effort\\" to control thinking behavior."}}'
     )
+    documented_disabled = (
+        'messages HTTP 400: {"error":{"message":"\\"thinking.type.disabled\\" is not supported for this '
+        'model. Use \\"thinking.type.adaptive\\" and \\"output_config.effort\\" to control thinking behavior."}}'
+    )
     gateway = (
         'messages HTTP 400: {"error":{"message":"claude-opus-5-5 requires adaptive thinking; omit thinking '
         'or use thinking.type=adaptive and output_config.effort"}}'
     )
-    assert extended_thinking_rejected(documented)
+    assert thinking_config_refused(documented_enabled)
-    assert extended_thinking_rejected(gateway)
+    assert thinking_config_refused(documented_disabled)
-    assert not extended_thinking_rejected('messages HTTP 400: {"error":{"message":"max_tokens: required"}}')
+    assert thinking_config_refused(gateway)
+    assert not thinking_config_refused('messages HTTP 400: {"error":{"message":"max_tokens: required"}}')

tests/test_agent/test_messages_dispatch.py

@@ -46,12 +46,19 @@ if TYPE_CHECKING:
     from collections.abc import AsyncIterator, Sequence
 
 
-# What the API answers a request that sends extended thinking to a model that
+# What the API answers a request whose thinking block it does not take (a gateway
-# takes adaptive thinking only (a gateway paraphrase of the documented wording).
+# paraphrase of the documented wording, which names adaptive thinking either way).
-_ADAPTIVE_REQUIRED_400 = (
+_THINKING_REFUSED_400: dict[str, str] = {
+    "enabled": (
         'messages HTTP 400: {"error":{"type":"error","message":"claude-opus-5-5 requires adaptive '
         'thinking; omit thinking or use thinking.type=adaptive and output_config.effort"}}'
-)
+    ),
+    "disabled": (
+        'messages HTTP 400: {"error":{"type":"error","message":"\\"thinking.type.disabled\\" is not '
+        'supported for this model. Use \\"thinking.type.adaptive\\" and \\"output_config.effort\\" to '
+        'control thinking behavior."}}'
+    ),
+}
 
 
 def _message_response(
@@ -83,12 +90,12 @@ class ScriptedMessagesClient:
     kind: str = "messages"
     calls: list[AnthropicMessagesParam]
     # Mirrors AnthropicClient: explicit thinking budget (0 = derive from effort),
-    # and the models the dispatch learned to send adaptive thinking to.
+    # and the thinking types the dispatch found the API refusing.
     thinking_budget_tokens: int
-    adaptive_thinking_models: set[str]
+    refused_thinking_types: dict[str, set[str]]
     _non_stream: list[AnthropicMessageResponse]
     _stream_events: list[list[AnthropicStreamEvent]]
-    _rejects_extended_thinking: bool
+    _refusals: frozenset[str]
     _error: str | None
 
     def __init__(
@@ -96,29 +103,29 @@ class ScriptedMessagesClient:
         *,
         non_stream: Sequence[AnthropicMessageResponse] | None = None,
         stream_events: Sequence[Sequence[AnthropicStreamEvent]] | None = None,
-        rejects_extended_thinking: bool = False,
+        refusals: Sequence[str] = (),
         error: str | None = None,
     ) -> None:
         self.calls = []
         self.thinking_budget_tokens = 0
-        self.adaptive_thinking_models = set()
+        self.refused_thinking_types = {}
         self._non_stream = list(non_stream or ())
         self._stream_events = [list(events) for events in (stream_events or ())]
-        self._rejects_extended_thinking = rejects_extended_thinking
+        self._refusals = frozenset(refusals)
         self._error = error
 
-    def needs_adaptive_thinking(self, model: str) -> bool:
+    def refuses_thinking_type(self, model: str, thinking_type: str) -> bool:
-        return model in self.adaptive_thinking_models
+        return thinking_type in self.refused_thinking_types.get(model, ())
 
-    def note_adaptive_thinking(self, model: str) -> None:
+    def note_refused_thinking_type(self, model: str, thinking_type: str) -> None:
-        self.adaptive_thinking_models.add(model)
+        self.refused_thinking_types.setdefault(model, set()).add(thinking_type)
 
     def _reject(self, param: AnthropicMessagesParam) -> None:
         """Raise what the API raises for a request it refuses to run."""
         if self._error is not None:
             raise RuntimeError(self._error)
-        if self._rejects_extended_thinking and param.thinking is not UNSET and param.thinking.type == "enabled":
+        if param.thinking is not UNSET and param.thinking.type in self._refusals:
-            raise RuntimeError(_ADAPTIVE_REQUIRED_400)
+            raise RuntimeError(_THINKING_REFUSED_400[param.thinking.type])
 
     @overload
     async def messages(
@@ -647,7 +654,7 @@ def _thinking_param(model: str = "claude-opus-5-5", effort: str = "high") -> Cha
 
 
 async def test_dispatch_messages_retries_with_adaptive_thinking() -> None:
-    client = ScriptedMessagesClient(non_stream=[_message_response(text="done")], rejects_extended_thinking=True)
+    client = ScriptedMessagesClient(non_stream=[_message_response(text="done")], refusals=("enabled",))
     result = await dispatch_messages(cast("Any", client), _thinking_param(), stream=False)
     assert result.choices[0].message.content == "done"
     assert len(client.calls) == 2
@@ -663,7 +670,7 @@ async def test_dispatch_messages_retries_with_adaptive_thinking() -> None:
 async def test_dispatch_messages_remembers_the_refusing_model() -> None:
     client = ScriptedMessagesClient(
         non_stream=[_message_response(text="one"), _message_response(text="two")],
-        rejects_extended_thinking=True,
+        refusals=("enabled",),
     )
     param = _thinking_param()
     _ = await dispatch_messages(cast("Any", client), param, stream=False)
@@ -672,6 +679,23 @@ async def test_dispatch_messages_remembers_the_refusing_model() -> None:
     assert [call.thinking.type for call in client.calls] == ["enabled", "adaptive", "adaptive"]
 
 
+async def test_dispatch_messages_drops_a_refused_disabled_block() -> None:
+    """A model whose thinking is always on answers ``disabled`` with a 400."""
+    client = ScriptedMessagesClient(
+        non_stream=[_message_response(text="one"), _message_response(text="two")],
+        refusals=("enabled", "disabled"),
+    )
+    param = _thinking_param(effort="none")
+    result = await dispatch_messages(cast("Any", client), param, stream=False)
+    assert result.choices[0].message.content == "one"
+    assert len(client.calls) == 2
+    assert client.calls[0].thinking.type == "disabled"
+    # No block at all: there is nothing left to ask such a model for.
+    assert client.calls[1].thinking is UNSET
+    _ = await dispatch_messages(cast("Any", client), param, stream=False)
+    assert client.calls[2].thinking is UNSET
+
+
 async def test_dispatch_messages_stream_retries_with_adaptive_thinking() -> None:
     events: list[AnthropicStreamEvent] = [
         AnthropicContentBlockStart(index=0, content_block=AnthropicRawContentBlock(type="text", text="")),
@@ -679,7 +703,7 @@ async def test_dispatch_messages_stream_retries_with_adaptive_thinking() -> None
         AnthropicMessageDelta(delta={"stop_reason": "end_turn"}, usage=AnthropicUsage(output_tokens=3)),
         AnthropicMessageStop(),
     ]
-    client = ScriptedMessagesClient(stream_events=[events], rejects_extended_thinking=True)
+    client = ScriptedMessagesClient(stream_events=[events], refusals=("enabled",))
     stream = await dispatch_messages(cast("Any", client), _thinking_param(), stream=True)
     chunks = [chunk async for chunk in cast("Any", stream)]
     texts = [c.choices[0].delta.content for c in chunks if c.choices and isinstance(c.choices[0].delta.content, str)]
@@ -695,8 +719,8 @@ async def test_dispatch_messages_does_not_retry_other_failures() -> None:
     assert len(client.calls) == 1
 
 
-async def test_dispatch_messages_does_not_retry_a_request_without_extended_thinking() -> None:
+async def test_dispatch_messages_does_not_retry_a_request_without_thinking() -> None:
-    client = ScriptedMessagesClient(error=_ADAPTIVE_REQUIRED_400)
+    client = ScriptedMessagesClient(error=_THINKING_REFUSED_400["enabled"])
     param = ChatCompletionsParam(model="claude-opus-5-5", messages=[UserChatMessage(content="hi")])
     with pytest.raises(RuntimeError, match="adaptive thinking"):
         _ = await dispatch_messages(cast("Any", client), param, stream=False)