ChatAnthropic rejects supported mid-conversation system messages

[sazuyakun]

Example Code

import { ChatAnthropic } from "@langchain/anthropic";
import { HumanMessage, SystemMessage } from "@langchain/core/messages";

const model = new ChatAnthropic({ model: "claude-opus-5" });

await model.invoke([
  new SystemMessage("You are a coding assistant."),
  new HumanMessage("Update this project."),
  new SystemMessage("Follow the project README."),
]);

Error Message

Error: System messages are only permitted as the first passed message.

Description

Anthropic supports role: "system" messages after a user turn on recent models, but ChatAnthropic rejects them before sending the request.

The first SystemMessage should continue to use the top-level system field. Later SystemMessage instances should remain in messages with role: "system".

Anthropic docs: https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages

Related Python issue: langchain-ai/langchain#38651

System Info

• @langchain/anthropic: 1.5.8
• @langchain/core: 1.2.9
• Node.js: 22.22.3
• pnpm: 10.30.3
• macOS: 26.5.1

[sazuyakun]

I ran into this limitation while building an agent with recent Claude models, including Opus 4.8 and Opus 5: ChatAnthropic rejected mid-conversation SystemMessage instances before sending the request.
I’ve opened #11473 with a focused fix and regression test.

Hunter Lovell (@hntrl) , when you have a chance, would you mind taking a look? Thanks!