Claude 4 "Interleaved Thinking"

[NN1985]

What happened?

Currently, Claude 4’s “thinking” blocks are enabled and working with both the Anthropic API and the Bedrock API.

However, Claude 4 introduces a new behavior for these blocks. Unlike previous versions, where thinking happened only at the beginning, Claude 4 can insert new blocks throughout the response, interleaving them with regular text and tool calls. This is referred to as "interleaved thinking."

Right now, LibreChat still handles Claude 4 thinking blocks the old Claude 3.7 way: it collects and displays them all together at the start of the message, stacking them into one big block. This leads to a problem: when Claude 4 generates new blocks during its response, they don’t appear in the correct place; they get grouped with the initial thinking block instead of appearing inline, where they’re meant to be.

This behavior doesn’t match Claude 4’s intended design and results in a broken or misleading display of responses in LibreChat.

Proposal: Update LibreChat to fully support the new interleaved block format used by Claude 4, to ensure accurate and modern response rendering, especially as Claude 4 becomes more widely used.

Version Information

Current Version.

Steps to Reproduce

Output a sufficiently long task to have Claude 4 trigger multiple thinking blocks.

What browsers are you seeing the problem on?

No response

Relevant log output

N/A

Screenshots

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

Interleaved thinking is a separate feature from Anthropic, it needs to be implemented, not a bug.