Checked other resources

• This is a feature request, not a bug report or usage question.
• I added a clear and descriptive title that summarizes the feature request.
• I used the GitHub search to find a similar feature request and didn't find it.
• I checked the LangChain documentation and API reference to see if this feature already exists.
• This is not related to the langchain-community package.

Package (Required)

• langchain
• langchain-openai
• langchain-anthropic
• langchain-classic
• langchain-core
• langchain-model-profiles
• langchain-tests
• langchain-text-splitters
• langchain-chroma
• langchain-deepseek
• langchain-exa
• langchain-fireworks
• langchain-groq
• langchain-huggingface
• langchain-mistralai
• langchain-nomic
• langchain-ollama
• langchain-openrouter
• langchain-perplexity
• langchain-qdrant
• langchain-xai
• Other / not sure / general

Feature Description

Hello! Anthropic has added support for adaptive thinking for Claude Opus 4.6 and Sonnet 4.6. I propose we update the chat_models.py file to take into account adaptive thinking wherever there is validation for thinking.

As an example we should change:

def _thinking_in_params(params: dict) -> bool:
    return params.get("thinking", {}).get("type") == "enabled"

to

def _thinking_in_params(params: dict) -> bool:
    return params.get("thinking", {}).get("type") in ("enabled", "adaptive")

This validation should be extended in all places applicable.

Optional: we could also add a comment that enabled is marked for deprecation and will be removed in the future.

Use Case

In our agentic use cases we have started using Opus 4.6 with adaptive thinking. For example:

llm = init_chat_model(
  model="anthropic:claude-opus-4.6-1m",
  temperature=1,
  max_tokens=128000,
  ...
  thinking={
    "type": "adaptive",
  },
)
...
response = llm.invoke([HumanMessage(content=...)])

We have noticed that in some cases that response.content is malformed and looks like this ['', {'type': 'thinking', ...}, "Actual response..."] and this translates in response.content_blocks only containing reasoning content. This is because the actual response is not a dict of type text.

When we change from adaptive thinking back to enabled it works.

We have started investigating this and this has lead us to look into the langchain repository and we found out that the Anthropic chat model only checks that thinking is enabled and not adaptive.

Proposed Solution

Wherever there is validation that thinking is enabled:

def _thinking_in_params(params: dict) -> bool:
    return params.get("thinking", {}).get("type") == "enabled"

we expand the validation to check that thinking is adaptive:

def _thinking_in_params(params: dict) -> bool:
    return params.get("thinking", {}).get("type") in ("enabled", "adaptive")

We also propose to add / extend unit tests where applicable.

Alternatives Considered

Using enabled instead of adaptive works around the issue, but adaptive is the recommended thinking mode going forward. enabled will also be deprecated in the future so it's not a long term solution.

Additional Context

Anthropic API docs on adaptive thinking: https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking