bug(@langchain/core or @langchain/google) Using `outputVersion: "v1"` inlines message content FOUR TIMES inside one `.toJSON()` output.

[vjancik]

Checked other resources

• This is a bug, not a usage question. For questions, please use the LangChain Forum (https://forum.langchain.com/).
• 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

/**
 * Dev script for testing Gemini's built-in code execution tool.
 * Invokes the model with a prompt that requires computation and saves the raw
 * AIMessage JSON to a timestamped file in this directory.
 */

import { HumanMessage, type MessageOutputVersion } from "@langchain/core/messages";
import { ChatGoogle } from "@langchain/google/node";

const MODEL = "gemini-2.5-flash";
const PROMPT = "Render the normal distribution of IQ scores for me";

async function runCodeExecution(outputVersion: MessageOutputVersion): Promise<void> {
  const llm = new ChatGoogle({
    model: MODEL,
    outputVersion,
  }).bindTools([
    {
      codeExecution: {},
    },
  ]);

  console.log(`Invoking model (outputVersion="${outputVersion}") with: "${PROMPT}"`);

  const result = await llm.invoke([new HumanMessage(PROMPT)]);

  const timestamp = new Date().toISOString().replace(/[:.]/g, "-");
  const outPath = `${import.meta.dir}/gemini-code-execution-${timestamp}-${outputVersion}.json`;

  await Bun.write(outPath, JSON.stringify(result.toJSON(), null, 2));

  console.log(`Result saved to: ${outPath}`);
}

// await runCodeExecution("v0");

await runCodeExecution("v1");

Error Message and Stack Trace (if applicable)

No response

Description

I'm attaching 3 example outputs of the previous script.

The issue becomes doubly problematic if the model output includes inlined data like images.

To spell out the problem, in outputVersion: "v1", the .toJSON() function saves the content of the AIMessage / AIMessageChunk
under:

1. .kwargs > .lc_kwargs > .lc_kwargs > .content
2. .kwargs > .lc_kwargs > .contentBlocks
3. .kwargs > .content
4. .kwargs > .content_blocks

As long as you are serializing messages directly into a DB store, even if you filter out inline base64 data, then whether you use .toJSON() or .toDict() or .mapChatMessagesToStoredMessages, the data is being inlined 4 times.

Now, there is a chance this is a @langchain/google specific bug, and somebody should try to reproduce this with other providers.

gemini-code-execution-2026-03-31T12-40-05-527Z-v1.json
gemini-code-execution-2026-03-31T12-37-09-729Z-v1.json
gemini-code-execution-2026-03-31T12-36-32-889Z-v0.json

System Info

Package: @langchain/google
Version: 0.1.9
Companion: @langchain/core 1.1.37 (peerDependency enforced)
Runtime: bun 1.3.11
Platform: Ubuntu 25.10 (Linux 6.17.0-14-generic)

[vjancik]

Ok, I explicitly set outputVersion: "v0" and now I'm still observing the double-triple nested lc_kwargs and content duplication.

[
  {
    "lc": 1,
    "type": "constructor",
    "id": [
      "langchain_core",
      "messages",
      "AIMessage"
    ],
    "kwargs": {
      "lc_serializable": true,
      "lc_kwargs": {
        "lc_serializable": true,
        "lc_kwargs": {
          "lc_serializable": true,
          "lc_kwargs": {
            "content": [

as seen here: #10567

That's serialized with a single toJSON() call on an AIMessage from .stream() of ChatGoogle with outputVersion: "v1"

Only thing I'm doing different is that I am using chunkA.concat(chunkB) and then casting the final chunk into an AIMessage as
new AIMessage({...chunk}). Is that what's causing this?

[vjancik]

Ok so if I do new AIMessage(chunk.lc_kwargs)
I got it back down to 2 nested lc_kwargs

[
  {
    "lc": 1,
    "type": "constructor",
    "id": [
      "langchain_core",
      "messages",
      "AIMessage"
    ],
    "kwargs": {
      "lc_serializable": true,
      "lc_kwargs": {
        "lc_serializable": true,
        "lc_kwargs": {
          "content": [
            {
              "type": "text",

It feels like .toJSON() isn't playing nice with lc_kwargs, it looks for kwargs and then dumps everything inside kwargs when it doesn't find it?

This is definitely a regression of some kind. It wasn't doing this before.

[vjancik]

Ok, I found all places where I was doing new AIMessage({ ...chunk }) or new AIMessageChunk({ ...chunk }). The nested lc_kwargs happens if you do that, and you have to do new AIMessageChunk(chunk.lc_kwargs) or new AIMessageChunk({ ...chunk.lc_kwargs }) instead. But the v1 LLM output -> .toJSON() still causes that, so I am guessing it's doing some rest spreading somewhere internally, which works with v0 output, but not with v1 output.