llama.cpp and its underlying ggml tensor library have reached version 1.0, with maintainers committing to make incompatible public-interface changes visible through major version numbers. The policy gives applications built on the local AI stack a clearer signal when an upgrade may break their integration.
Under the new semantic-versioning rules, a breaking change to a public C API triggers a major release. Backward-compatible features, new model or hardware support and backend updates use a minor release, while compatible bug fixes use a patch. The promise applies to public headers rather than every internal implementation detail.
Distribution is also expanding beyond source builds. The project lists planned or available artifacts for Linux, macOS, Windows, Android and iOS, alongside channels including Docker, Debian, Homebrew, conda-forge, MacPorts and Winget. Dynamic backends let one build load hardware-specific implementations at runtime.
The maintainers say llama.cpp now supports more than 150 model architecture variants and mainstream speculative-decoding methods. Broader continuous integration is being added with self-hosted hardware from several chip vendors. Version 1.0 does not mean the software has stopped changing; it means downstream developers should receive more predictable notice when its supported interfaces do.