llama.cpp Releases Build b10625 with Qwen3-Coder Chat Workaround
llama.cpp has released build b10625, which introduces a specific workaround for the Qwen3-Coder model when running in chat mode. This update ensures better compatibility and smoother execution for developers utilizing the latest Qwen3-Coder models within the llama.cpp ecosystem. The release includes pre-built binaries for various platforms like Windows, Linux, macOS, and Android, though the macOS Apple Silicon build with KleidiAI support remains disabled.
## BACKGROUND
llama.cpp is an open-source LLM inference engine written in C/C++ designed for efficient local execution of models. Qwen3-Coder is a code-focused large language model developed by the Qwen team, optimized for programming tasks and agentic workflows.