invalid_request_error when using Code Interpreter with Claude extended thinking enabled

[agluo]

📱 Client Type

Web (Desktop Browser)

💻 Operating System

Windows

📦 Deployment Platform

Self hosting Docker

🔧 Deployment Mode

server db (lobe-chat-database image)

📌 Version

v2.0.0-next.147

🌐 Browser

Edge

🐛 Bug Description

Description

Getting an invalid_request_error from Anthropic API when trying to execute Python code using the Code Interpreter with Claude Haiku (claude-haiku-4-5-20251001).

Error Message

{
  "error": {
    "type": "invalid_request_error",
    "message": "messages.1.content.0.type: Expected `thinking` or `redacted_thinking`, but found `text`. When `thinking` is enabled, a final `assistant` message must start with a thinking block (preceeding the lastmost set of `tool_use` and `tool_result` blocks). We recommend you include thinking blocks from previous turns. To avoid this requirement, disable `thinking`. Please consult our documentation at https://docs.claude.com/en/docs/build-with-claude/extended-thinking"
  },
  "endpoint": "https://api.anthropic.com",
  "provider": "anthropic"
}

顺带一提，其他插件也会报错

📷 Recurrence Steps

1. Use Claude Haiku model (claude-haiku-4-5-20251001) with extended thinking enabled
2. Request to write and execute a Python program via Code Interpreter
3. Example prompt: "写一段python程序计算第15个素数，并且用插件运行" (Write a Python program to calculate the 15th prime number and run it with the plugin)
4. Error occurs when trying to execute

🚦 Expected Behavior

The Code Interpreter should execute the Python code successfully without API errors.

📝 Additional Information

🛠️ Willing to Submit a PR?

No, but I am happy to help test the fix

✅ Validations

• Read the docs.
• Check that there isn't already an issue that reports the same bug to avoid creating a duplicate.
• Make sure this is a LobeChat issue and not a third-party library or provider issue.
• Check that this is a concrete bug. For Q&A, please use GitHub Discussions or join our Discord Server.

[lobehubbot]

@arvinxx @sxjeru - This is a tool calling issue with Claude's extended thinking feature. The API expects thinking blocks in the message format when extended thinking is enabled, but LobeChat is not handling this correctly for tool use scenarios. Please coordinate on the fix.

[dosubot]

This error happens because Anthropic's Claude models require every assistant message to start with a thinking or redacted_thinking block when extended thinking is enabled. LobeChat tries to include these blocks, but if the previous message was a plugin/tool call (like Code Interpreter) and didn't produce a reasoning signature, the next assistant message may start with a plain text block instead. This violates Claude's contract and triggers the invalid_request_error you see source.

Recent updates have improved message handling, but edge cases remain—especially after plugin/tool calls. A fix is in progress to ensure a thinking block is always present when required, even for plugin-only responses source.

For now, you can try these workarounds:

• Retry the message once; this often lets the conversation proceed normally source.
• Disable the history limit in settings, which can help avoid message structure issues source.
• If you need reliability, consider temporarily disabling extended thinking until a permanent fix lands.

Keep an eye on new releases—this area is actively being improved. If you’re able to test, feedback on edge cases is helpful for the maintainers.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}