Qwen3 Next 80B fails to reuse cache

[di-copilot]

Which version of LM Studio?
0.3.33 (build 1)

Which operating system?
Windows 11

What is the bug?
It always fails to remove tokens from cache and clears the entire cache, making the model do the entire decoding from the beginning

Screenshots

Logs
Add any relevant logs.

To Reproduce

1. Download Qwen3 Next and Set Config mentioned in the screenshot
2. Open VS Code, use any open-source Agent like Continue Dev to send a message to LM Studio
3. Check logs for clear entire cache

NOTE: It does not happens when using lm studio's built in chat