Name and Version

literally head

Device 0: NVIDIA GeForce RTX 3080, compute capability 8.6, VMM: yes
version: 6134 (be48528)
built with MSVC 19.41.34120.0 for x64

also tested at the office on ubuntu, on head

Operating systems

Linux

Which llama.cpp modules do you know to be affected?

llama-server

Command line

` -fa -ngl 999999 -ctk q4_0 -ctv q4_0 -c 128000  --jinja -m Qwen3-4B-Q5_K_S.gguf --port 2483 --slots`

Problem description & steps to reproduce

If you call chat.completions.create on a reasoning model for example here Qwen3 with tool_choice: "required"

It will refuse to do the reasoning because of the grammar implemented.

First Bad Commit

No response

Relevant log output