langchain - ✅(Solved) Fix openai model: tool_choice="required" not compatible with thinking models (qwen3.5-plus) [1 pull requests, 3 comments, 3 participants]

langchain2026-03-20 07:00:04

ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

GitHub issue URL

Helpful · Quick feedback

GitHub stats

langchain-ai/langchain#36120•Fetched 2026-04-08 01:03:30

View on GitHub

Comments

Participants

Timeline

Reactions

Author

Participants

Timeline (top)

labeled ×3commented ×2cross-referenced ×1issue_type_added ×1

Error Message

System Info

langchain-openai version: 0.1.20
langchain version: 1.2.x
model: qwen3.5-plus (Tongyi)
base_url: Dashscope compatible OpenAI protocol

Issue Description

When using thinking models (like qwen3.5-plus, deepseek-r1) with ChatOpenAI / init_chat_model, the LangChain agent automatically sets tool_choice="required", which causes the API to return a 400 error:

Fix Action

Fix / Workaround

This is a bug, not a usage question.
I added a clear and descriptive title that summarizes this issue.
I used the GitHub search to find a similar question and didn't find it.
I am sure that this is a bug in LangChain rather than my code.
The bug is not resolved by updating to the latest stable version of LangChain (or the specific integration package).
This is not related to the langchain-community package.
I posted a self-contained, minimal, reproducible example. A maintainer can copy it and run it AS IS.

PR fix notes

PR #36135: fix(openai): downgrade tool_choice='required' to 'auto' for reasoning models

Repository: langchain-ai/langchain
Author: majiayu000
State: closed | merged: False
Link: https://github.com/langchain-ai/langchain/pull/36135

Description (problem / solution / changelog)

Fixes #36120

Problem

When using reasoning/thinking models (o1, o3, gpt-5, qwen3.5-plus, etc.) with LangChain agents, the agent factory auto-sets tool_choice='any' which gets normalized to tool_choice='required' in ChatOpenAI.bind_tools(). Reasoning models reject tool_choice='required' with HTTP 400 errors.

Changes

In ChatOpenAI.bind_tools(), after the tool_choice normalization block converts 'any' → 'required' and True → 'required', added a check: if the resulting tool_choice is 'required' AND the model is a reasoning model, downgrade to 'auto' with a warning.

A reasoning model is detected by:

self.profile has reasoning_output=True, OR
self.reasoning_effort is set, OR
self.reasoning is set

Test plan

Added unit tests verifying reasoning models get 'auto' instead of 'required'
Added regression tests verifying non-reasoning models still get 'required'
Added tests verifying explicit tool_choice='auto'/'none' are unaffected
All 324 existing tests pass

Disclaimer

This PR was developed with AI agent assistance.

Changed files

libs/partners/openai/langchain_openai/chat_models/base.py (modified, +21/-0)
libs/partners/openai/tests/unit_tests/chat_models/test_base.py (modified, +86/-0)

Code Example

### System Info
- langchain-openai version: 0.1.20
- langchain version: 1.2.x
- model: qwen3.5-plus (Tongyi)
- base_url: Dashscope compatible OpenAI protocol

### Issue Description
When using **thinking models** (like qwen3.5-plus, deepseek-r1) with `ChatOpenAI / init_chat_model`, the LangChain agent automatically sets `tool_choice="required"`, which causes the API to return a 400 error:

---

Thinking models **do NOT support `tool_choice="required"`**, only "auto" or "none".

But in `langchain/agents/factory.py`, `_execute_model_sync` forces `tool_choice="required"`, which breaks all thinking models.

### Error Log

RAW_BUFFERClick to expand / collapse

Checked other resources

This is a bug, not a usage question.
I added a clear and descriptive title that summarizes this issue.
I used the GitHub search to find a similar question and didn't find it.
I am sure that this is a bug in LangChain rather than my code.
The bug is not resolved by updating to the latest stable version of LangChain (or the specific integration package).
This is not related to the langchain-community package.
I posted a self-contained, minimal, reproducible example. A maintainer can copy it and run it AS IS.

Package (Required)

Related Issues / PRs

System Info

langchain-openai version: 0.1.20
langchain version: 1.2.x
model: qwen3.5-plus (Tongyi)
base_url: Dashscope compatible OpenAI protocol

Issue Description

Reproduction Steps / Example Code (Python)

### System Info
- langchain-openai version: 0.1.20
- langchain version: 1.2.x
- model: qwen3.5-plus (Tongyi)
- base_url: Dashscope compatible OpenAI protocol

### Issue Description
When using **thinking models** (like qwen3.5-plus, deepseek-r1) with `ChatOpenAI / init_chat_model`, the LangChain agent automatically sets `tool_choice="required"`, which causes the API to return a 400 error:

Error Message and Stack Trace (if applicable)

Thinking models **do NOT support `tool_choice="required"`**, only "auto" or "none".

But in `langchain/agents/factory.py`, `_execute_model_sync` forces `tool_choice="required"`, which breaks all thinking models.

### Error Log

Description

Suggestion

Add a check for thinking models, or allow users to force override tool_choice without modifying source code.

Or make tool_choice="auto" default for agent model calls.

System Info

Suggestion

Add a check for thinking models, or allow users to force override tool_choice without modifying source code.

Or make tool_choice="auto" default for agent model calls.

extent analysis

Fix Plan

To resolve the issue, we need to modify the langchain library to handle thinking models correctly. Here are the steps:

Modify the _execute_model_sync function in langchain/agents/factory.py to allow for a default tool_choice of "auto" or to override the tool_choice parameter.
Add a check for thinking models and set tool_choice to "auto" by default for these models.

Example Code

# In langchain/agents/factory.py

def _execute_model_sync(self, model, prompt, tool_choice="auto"):
    # ... existing code ...

    if model.startswith("thinking-") or model in ["qwen3.5-plus", "deepseek-r1"]:
        tool_choice = "auto"

    # ... existing code ...

Alternatively, you can add an optional tool_choice parameter to the init_chat_model function:

# In langchain/agents/factory.py

def init_chat_model(self, model, tool_choice="auto"):
    # ... existing code ...

    self._execute_model_sync(model, prompt, tool_choice=tool_choice)

Then, when initializing the chat model, you can specify the tool_choice parameter:

# Example usage
agent = ChatOpenAI()
agent.init_chat_model("qwen3.5-plus", tool_choice="auto")

Verification

To verify that the fix worked, you can test the init_chat_model function with a thinking model and check that the API returns a successful response:

# Example test
agent = ChatOpenAI()
agent.init_chat_model("qwen3.5-plus", tool_choice="auto")
response = agent.get_response("Hello, world!")
print(response)  # Should print a successful response

Vote matrix · Quick signals

Works

Did the solution work? Tap to confirm.

Easy Fix

Was it a quick fix?

Time Saver

Did it save you time?

Blocking

Was it severely blocking?

Common Issue

Are others likely hitting this too?

Flaky / Intermittent

Is it intermittent?

Verified / Reproducible

Can you reproduce it reliably?

#api #ssr #installation #tensor shape #autograd error #search optimization #API routing #API middleware

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

langchain - ✅(Solved) Fix openai model: tool_choice="required" not compatible with thinking models (qwen3.5-plus) [1 pull requests, 3 comments, 3 participants]

Recommended Tools

GitHub issue graph ai analysis

Error Message

System Info

Issue Description

Fix Action

Fix / Workaround

PR fix notes

PR #36135: fix(openai): downgrade tool_choice='required' to 'auto' for reasoning models

Description (problem / solution / changelog)

Problem

Changes

Test plan

Disclaimer

Changed files

Code Example

Checked other resources

Package (Required)

Related Issues / PRs

System Info

Issue Description

Reproduction Steps / Example Code (Python)

Error Message and Stack Trace (if applicable)

Description

Suggestion

System Info

Suggestion

extent analysis

Fix Plan

Example Code

Verification

Still need to ship something?

RELATED_DISCOVERY

TRENDING