llama.cpp and ggml reach version 1.0.0 with a stable API and semantic versioning

The team behind llama.cpp announced version 1.0.0 of both llama.cpp and its underlying tensor library, ggml, on November 13, 2026, in a post on the Hugging Face blog. llama.cpp is a widely used open-source engine for running language models locally, from laptops to servers, and the team frames the release as a commitment to the many apps built on top of it.

What's new

• Stable interfaces: from now on, the team says, incompatible changes to the API or ABI of llama.cpp and ggml will be marked with a new major version.
• Versioning rules: under the rules published for ggml-org projects, a breaking change to the public C API raises the major number, new features, backends, model support or API additions raise the minor number, and bug fixes without API changes raise the patch number.
• Distribution: the post lists the channels where the projects ship, including pre-release builds for Linux, macOS, Windows, Android and iOS, plus Docker, Debian, Homebrew, conda-forge, MacPorts, Winget and an XCFramework for Apple platforms.
• Coverage: according to the team, llama.cpp now supports more than 150 model architecture variants as well as the mainstream speculative decoding methods.
• Testing: the team says it is expanding its CI using Hugging Face infrastructure, with NVIDIA, AMD, Qualcomm, Arm, IBM and Intel supplying hardware for self-hosted test runners.

The project is about 3.5 years old, according to the post.

Why it matters

Analysis: the team says many downstream applications already depend on its libraries, and without a version contract an update could break them with little warning. A clear version contract makes it safer for app developers to upgrade, and it should mean fewer broken builds for people who run models on their own machines. For users of local AI tools, the practical effect is likely to be steadier updates rather than new features.