llama.cpp b10996 Fixes Reasoning Tag Closure for Qwen3-Coder
llama.cpp release b10996 introduces a fix (PR #28869) that forces a closing `\n</think>` tag when the reasoning budget ends during Qwen3-Coder model inference. Without proper closing tags, chat interfaces and parsers can misinterpret model output or fail to separate chain-of-thought reasoning from the final answer. This update ensures consistent output formatting when reasoning token limits are reached. The patch specifically targets token stream output behavior under PR #28869 for Qwen3-Coder reasoning budget termination. Binaries for this automated release tag are distributed across macOS, Linux, Windows, Android, CUDA, Vulkan, and SYCL environments.
## BACKGROUND
llama.cpp is a high-performance open-source C/C++ engine used for running Large Language Models locally across diverse hardware platforms. Modern reasoning models use explicit formatting tags like `<think>` and `</think>` to encapsulate internal chain-of-thought steps before displaying the final response, controlled by a reasoning budget limit.