Coding & agents New
llama.cpp reaches version 1.0.0 and promises semantic versioning for its local-inference libraries
Published:
The ggml-org project released llama.cpp v1.0.0 and ggml v1.0.0 on 13 November. In a Hugging Face blog post, the maintainers say both libraries now follow semantic versioning, so breaking changes to the public C API will bring a major-version bump.
The release notes say v1.0.0 removes 35 deprecated functions, adds FP8 key-value caches and a host-memory cache for mixture-of-experts weights, and adds support for new models including Kimi-K3.1-Pro. It also hardens llama-server against SSRF, CORS and DNS-rebinding attacks and adds an OpenAI-compatible speech endpoint.
Caveat. Developers upgrading from earlier builds must adapt to the removed deprecated functions. Performance gains are the project's own description, not independently measured here.
Sources