The open-source llama.cpp 1.0 project, the foundation of thousands of local artificial intelligence setups, reached version 1.0.0 on November 13, 2026. Alongside it, the ggml tensor library also hit v1.0.0. The main promise is the formal adoption of semantic versioning for the public C API.
What changes with llama.cpp 1.0
According to the announcement on the Hugging Face blog, from now on any change that breaks the public C API will require a major version bump. New compatible features enter as minor, and bug fixes as patch. The ggml-org team says the goal is to give predictability to those building tools on top of the library.
The release removes 35 deprecated functions from the public API and promotes the NextN speculative decoding API to the main llama.h header. Technical additions include FP8 KV caches, a host-memory MoE expert cache with CPU usage control, and support for more than 150 model architectures.
Why it matters for local AI users
llama.cpp is the engine behind tools such as Ollama, LM Studio and dozens of frontends that let people run large models on laptops, desktops and even modest devices. API stability reduces the risk of unexpected breaks in future updates and makes maintenance of bindings and applications easier.
The llama-server also received security improvements, including protections against SSRF, CORS issues and DNS rebinding, plus a new speech endpoint.
Sources
This content was produced with the assistance of artificial intelligence and editorially reviewed.
Por Grok
