llama.cpp is a free, open-source engine for running AI language models on your own computer or phone, and it's the foundation under a large share of local-AI apps. On November 13, 2026 it reached version 1.0, which means a stable, versioned API so updates stop breaking the apps built on it.
llama.cpp at a glance
Price | Free, open source |
|---|---|
Models | 150+ architecture variants |
Platforms | Linux, macOS, Windows, Android, iOS |
Install via | Homebrew, Winget, Docker, Debian, conda-forge, MacPorts, XCFramework |
Maintained by | ggml-org |
What llama.cpp does
llama.cpp loads an open-weight model and runs it efficiently on ordinary hardware, from laptops to phones to GPU servers. Each hardware backend can be compiled as a separate library and loaded at runtime. It supports more than 150 model architecture variants and all mainstream speculative decoding methods, which speed up generation.
What changed in version 1.0
The libllama and libggml libraries were released as v1.0.0.
Semantic versioning: a major bump now means a breaking change to the public C API, minor means new features or model support, and patch means bug fixes.
More investment in CI to support faster development.
What you can do with it
Chat with open models offline, with nothing leaving your machine.
Power your own app or agent with a local model behind an API.
Run models on Android or iOS devices.
Our take
Version 1.0 is a quiet but important milestone. llama.cpp has always been fast-moving, which made it painful to build on; a stable, versioned API makes it a dependable foundation for products, not just hobby projects. If you want private AI on your own hardware, this is still the place to start.
llama.cpp alternatives
Edge0-145B: a large open model designed to run on one GPU.
ZeroClaw: a self-hosted agent that can use local models.
Frequently asked questions
How do I install llama.cpp?
Use a package manager such as Homebrew (macOS, Linux), Winget (Windows) or Docker, or download builds from the GitHub releases page.
Does it need a GPU?
No, it runs on CPUs too, though a GPU backend is much faster for larger models.
As featured in World of AI
DeepSeek puts its whole app into the API (Nov 13, 2026)
Facts checked against the llama.cpp 1.0 announcement on November 13, 2026.
Want four tools like this every weeknight? Subscribe to World of AI for free. Browse more in our AI Tools directory.