Official releaseOpen Models / On-prem AI

llama.cpp Releases v1.0.0: Major Versions Now Signal API and ABI Breaks

llama.cpp and ggml released libraries v1.0.0 on November 13, 2026, committing to signal API and ABI breaks with major-version bumps. Teams embedding local inference get more predictable upgrades.

Key answer

llama.cpp v1.0.0 is the library release announced by ggml-org on November 13, 2026, alongside ggml v1.0.0. Both now follow semantic versioning: only a breaking change to the public C API bumps the major version. It matters because downstream applications can judge upgrade compatibility from the version number.

The ggml-org team announced libraries v1.0.0 for llama.cpp and ggml on the Hugging Face blog on November 13, 2026. The post says that in the three and a half years since the project began, many downstream applications have come to depend on its library interfaces, so a flexible but stable API is essential. Going forward, both projects will reflect API and ABI incompatibilities with major-version bumps.

The versioning rules follow semantic versioning and apply to all ggml-org projects that use it: a breaking change to the public C API under the include directory is a major bump; a backward-compatible feature, backend update, model support, or API addition is a minor bump; and a bug fix with no API change is a patch. The corresponding libraries are libllama and libggml.

For distribution, the post lists pre-release artifacts (Linux, macOS, Windows, Android, iOS), Docker, Debian, Homebrew, conda-forge, MacPorts, Winget, and XCFramework. The team says it wants to expand the ways its software reaches developers and end users.

On coverage, the post says llama.cpp currently supports more than 150 architecture variants and all mainstream speculative decoding methods. The underlying ggml engine supports different hardware through dedicated backends, each of which can be compiled as a dynamic library and loaded at runtime. Continuous integration runs on Hugging Face infrastructure, with self-hosted runners backed by hardware support from NVIDIA, AMD, Qualcomm, Arm, IBM, and Intel.

The post also addresses how the project is developed: the team says it integrates LLMs into its development cycle thoughtfully, and names maintaining code quality, staying in control of the architecture, and understanding implementation details as its top priorities.

For enterprises that embed llama.cpp in desktop tools, edge devices, or internal services, three practical steps follow. Pin the dependency to an explicit version and treat a major bump as the signal for regression testing. Confirm that you use the API in the public headers and do not depend on internal interfaces with no stability commitment. And check whether an official distribution channel can replace your own builds, which reduces the cost of maintaining build pipelines across platforms.

X Cube view

A local inference library making an explicit commitment on interface stability shows it is now treated as a foundation for other products. When planning on-prem or edge AI, enterprises can add the inference engine's versioning policy to their selection criteria, alongside model licensing and hardware support.