AzureChatOpenAI does not support reasoning summaries

[tcoca27]

Checked other resources

• I added a very descriptive title to this issue.
• I searched the LangChain.js documentation with the integrated search.
• 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.js rather than my code.
• The bug is not resolved by updating to the latest stable version of LangChain (or the specific integration package).

Example Code

const model = new AzureChatOpenAI({
        modelName: 'o4-mini',
        streaming: true,
        azureOpenAIApiKey: process.env.AZURE_OPENAI_API_KEY_41,
        azureOpenAIApiInstanceName: process.env.AZURE_OPENAI_API_INSTANCE,
        azureOpenAIApiDeploymentName: 'o4-mini',
        azureOpenAIApiVersion: '2025-04-01-preview',
        reasoningEffort: 'low',
        __includeRawResponse: true,
        timeout: 220000,
        reasoning: {
          summary: 'auto',
        },
      });

// ...

const pipedLlm = ragPrompt.pipe(model);

const response = await pipedLlm.invoke({
//...
});

Error Message and Stack Trace (if applicable)

Troubleshooting URL: https://js.langchain.com/docs/troubleshooting/errors/MODEL_NOT_FOUND/
@web:dev: 
@web:dev:     at APIError.generate (webpack-internal:///(rsc)/../../node_modules/.pnpm/openai@4.104.0_encoding@0.1.13_ws@8.18.1_zod@3.24.3/node_modules/openai/error.mjs:69:20)
@web:dev:     at AzureOpenAI.makeStatusError (webpack-internal:///(rsc)/../../node_modules/.pnpm/openai@4.104.0_encoding@0.1.13_ws@8.18.1_zod@3.24.3/node_modules/openai/core.mjs:333:65)
@web:dev:     at AzureOpenAI.makeRequest (webpack-internal:///(rsc)/../../node_modules/.pnpm/openai@4.104.0_encoding@0.1.13_ws@8.18.1_zod@3.24.3/node_modules/openai/core.mjs:377:30)
@web:dev:     at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
@web:dev:     at async eval (webpack-internal:///(rsc)/../../node_modules/.pnpm/@langchain+openai@0.5.12_@langchain+core@0.3.57_openai@4.67.3_encoding@0.1.13_zod@3.24.3___encoding@0.1.13_ws@8.18.1/node_modules/@langchain/openai/dist/chat_models.js:2370:24)
@web:dev:     at async RetryOperation.eval [as _fn] (webpack-internal:///(rsc)/../../node_modules/.pnpm/p-retry@4.6.2/node_modules/p-retry/index.js:50:12) {
@web:dev:   status: 404,
@web:dev:   headers: {
@web:dev:     'apim-request-id': '2863c8c1-d34a-4cf1-9868-45a5866f3286',
@web:dev:     'content-length': '56',
@web:dev:     'content-type': 'application/json',
@web:dev:     date: 'Fri, 13 Jun 2025 06:42:15 GMT',
@web:dev:     'strict-transport-security': 'max-age=31536000; includeSubDomains; preload',
@web:dev:     'x-content-type-options': 'nosniff'
@web:dev:   },
@web:dev:   request_id: undefined,
@web:dev:   error: { code: '404', message: 'Resource not found' },
@web:dev:   code: '404',
@web:dev:   param: undefined,
@web:dev:   type: undefined,
@web:dev:   lc_error_code: 'MODEL_NOT_FOUND',
@web:dev:   attemptNumber: 1,
@web:dev:   retriesLeft: 6,
@web:dev:   pregelTaskId: '3923942f-e3cd-557f-b379-b87d914c76ea'
@web:dev: }

Description

AzureChatOpenAI throws a 404 error when trying to invoke the model with reasoning summaries enabled. What I'm trying to achieve is to stream thinking tokens and I followed the implementation here.

• If I remove the reasoning parameter in the model instantiation, I get a response, but it is missing reasoning tokens.
• this is reportedly fixed in the python package by passing in use_responses_api=True on instantiation here . This parameter is not available in the js library

System Info

"@langchain/aws": "^0.1.1",
"@langchain/core": "^0.3.57",
"@langchain/google-webauth": "^0.2.9",
"@langchain/groq": "^0.1.2",
"@langchain/langgraph": "^0.2.45",
"@langchain/openai": "^0.5.12",
"@langchain/pinecone": "^0.1.1",
"@langchain/textsplitters": "^0.1.0",
"langchain": "^0.3.2",
"langsmith": "^0.2.10",

[mauriciocirelli]

I am having the same issue...
Trying to use o3-pro model with AzureChatOpenAI and getting the MODEL_NOT_FOUND error.
Using any parameter like reasoning object or reasoningEffort causes this issue.
Also tried to use useResponsesApi flag, but it is not available in the latest packages.