llama.cpp and ggml Reach Version 1.0.0 with 150+ Model Architectures
Creator Georgi Gerganov announced the 1.0.0 release of llama.cpp and the ggml machine learning engine, introducing clear API compatibility rules and supporting over 150 model architectures. The update establishes formal semantic versioning for libllama and libggml to better manage breaking changes. This major milestone brings stability and predictable versioning to the foundational backend of the local AI ecosystem, making it easier for downstream software and developers to build on top of it. As the de facto standard for running large language models locally, formalizing API compatibility ensures long-term reliability for tools relying on these libraries. The update formalizes versioning rules where major version bumps signal breaking C API changes, minor updates include backward-compatible features, and patch releases handle bug fixes. Furthermore, the project expanded its CI infrastructure using Hugging Face tools and enhanced cross-platform hardware support across major vendors like NVIDIA, AMD, and Intel.
## BACKGROUND
llama.cpp is a widely used open-source C/C++ software library that performs inference on large language models directly on local hardware without requiring cloud connectivity. It is co-developed alongside ggml, a general-purpose tensor library that supports efficient CPU and GPU computations, quantization, and cross-platform execution.