Skip to content

AI chat follow-up messages fail with newer Claude models (HTTP 400)

Hi,

I'm using FMRTE 27.0.0-build3 with Football Manager 2027 and my own Anthropic API key.

I've run into an issue with the AI chat feature when using newer Claude models, such as Claude Opus 5.5, Sonnet 5.5, and Fable 5.1.

The first message in a new chat works perfectly, but every follow-up message in the same conversation fails with an HTTP 400 error. Starting a new chat allows me to ask another question, but the problem happens again as soon as I try to continue the conversation.

Error message:

messages.1.content.0: Invalid signature in thinking block. The block is bound to a different conversation. Remove the block, or set thinking.block_binding.prefix_mismatch_behavior to "drop_block". That setting requires the thinking-binding-controls-2026-08-01 value in the anthropic-beta header. The system prompt differs from the one this block was created with.

The API key itself works fine, so this appears to be related to how the conversation history is handled rather than an authentication issue.

I looked into Anthropic's official documentation and found that newer Claude models validate previous thinking blocks against the original conversation context, including the system prompt, tools, and earlier messages.

The error specifically indicates that the system prompt has changed between requests. I suspect FMRTE may be updating the system prompt with current game data while also reusing previous assistant responses containing thinking blocks.

I'm not sure how FMRTE handles this internally, but Anthropic documents a few possible solutions:

1. Keep the conversation context consistent: Keep the top-level system prompt unchanged throughout the conversation, and append updated game data as new messages instead of modifying the original prompt.

2. Remove invalidated thinking blocks: If the conversation context needs to change, remove the affected "thinking" block and all subsequent "thinking" and "redacted_thinking" blocks before resending the history, while preserving the other message content.

3. Use Anthropic's compatibility setting: Add the request header:

"anthropic-beta: thinking-binding-controls-2026-08-01"

And configure:

"thinking": {

"type": "adaptive",

"block_binding": {

"prefix_mismatch_behavior": "drop_block"

}

}

This allows the API to discard incompatible thinking blocks instead of rejecting the request.

The first approach would probably be the best long-term solution, as it preserves earlier reasoning and allows prompt caching to work properly.

According to Anthropic's documentation, this validation is enforced by default for accounts created on or after August 31, 2026, on the affected models, which may explain why some users experience the issue while others don't.

For now, my workaround is to start a new chat for every question or switch to Claude Opus 5 / Sonnet 5, which don't enforce this particular check.

Official Anthropic documentation:

Troubleshooting thinking (exact HTTP 400 error and recommended fixes):

https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting

Thinking (conversation history and thinking block preservation):

https://platform.claude.com/docs/en/build-with-claude/thinking

Claude API errors (thinking block bound to a different conversation):

https://platform.claude.com/docs/en/api/errors

Would it be possible to look into this for a future FMRTE update? It would be great to have multi-turn AI conversations working with the newer Claude models.

My setup:

- FMRTE: 27.0.0-build3

- Game: Football Manager 2027

- AI provider: Anthropic API

- Model: Claude Opus 5.5, Sonnet 5.5, and Fable 5.1

Thanks for your work on FMRTE!

Featured Replies

  • Author
19 hours ago, BraCa said:

Can you try with the latest version (build 5)?

Thanks

Thank you so much!

Build 5 is working perfectly now.

I really appreciate your help!

  • Author
19 hours ago, BraCa said:

Can you try with the latest version (build 5)?

Thanks

Oh,sorry! It's actually my first time posting here's I'm not really sure how things work yet.

I hope haven't posted this in wrong place 😅

Guest
This topic is now closed to further replies.

Account

Navigation

Search

Search

Configure browser push notifications

Chrome (Android)
  1. Tap the lock icon next to the address bar.
  2. Tap Permissions → Notifications.
  3. Adjust your preference.
Chrome (Desktop)
  1. Click the padlock icon in the address bar.
  2. Select Site settings.
  3. Find Notifications and adjust your preference.