llama.cpp reaches version 1.0: what changes for local AI

On November 13, 2026, the ggml-org team released llama.cpp 1.0.0, together with version 1.0.0 of the ggml library it is built on. llama.cpp is an open-source project that lets you run large language models (LLMs) directly on your computer, without sending your data to an external service.

After about three and a half years of development, it's the project's first "major" version. The announcement, published on the Hugging Face blog, presents it as a stronger foundation for the whole local AI ecosystem.

What is llama.cpp, in simple terms?

It's an engine: the piece of software that loads a model and makes it answer using your PC's processor or graphics card. It works with model files in the GGUF format, which you'll find in large numbers on Hugging Face. Developers and enthusiasts use it directly, and several simpler apps include it under the hood.

If running AI on your own computer is new to you, start with our guide running AI locally: what it is, when it makes sense and what you need (in Italian).

What does version 1.0 bring?

The main change isn't a feature but a promise of stability. From now on the project follows semantic versioning, a clear rule for numbering releases:

Number that changesWhat it means
Major (1.x.x → 2.0.0)Changes that break compatibility with programs using the library
Minor (1.0 → 1.1)New features, newly supported models, backend updates, without breaking anything
Patch (1.0.0 → 1.0.1)Bug fixes only

For anyone building tools on top of llama.cpp, this means knowing in advance when an update requires changing their own code. According to the announcement, also:

  • the project supports more than 150 model architecture variants;
  • ggml backends (the modules for different graphics cards and processors) can be loaded as separate libraries, only when needed;
  • llama.cpp can be installed with the most common package managers: Winget on Windows, Homebrew and MacPorts on Mac, Debian, Docker and conda-forge, plus prebuilt files for Linux, macOS, Windows, Android and iOS;
  • hardware support is credited to NVIDIA, AMD, Qualcomm, Arm, IBM and Intel.

What's in the release notes?

The official 1.0.0 release page on GitHub also lists more technical changes:

  • Server security: two sensitive options (unauthenticated tools and access to private addresses for media files) are now off by default and must be enabled manually; the Hugging Face token is sent only to Hugging Face hosts; several vulnerabilities were fixed.
  • Memory: an FP8 cache that uses less memory during long conversations, and a cache for MoE models (models made of several "experts") that can use system RAM.
  • New supported models, including Kimi-K3.1-Pro and Ripple-65B-Preview.
  • A new OpenAI-compatible text-to-speech endpoint and a redesigned web interface, with chat groups and better phone use.
  • For developers: 35 deprecated functions and some outdated options were removed.

What changes for people at work?

If you run AI locally to keep company documents and data private, version 1.0 is good news for two main reasons. The first is security: the riskiest server settings are now off until you turn them on, an important detail if the server runs on a company network. The second is predictability: tools built on llama.cpp will be able to update with fewer surprises.

If you use ready-made apps, you don't need to do anything: the improvements will arrive when their developers update the engine. If you use llama.cpp directly, read the release notes before updating, because some options and functions have been removed.

Frequently asked questions

Is llama.cpp free?

Yes, it's an open-source project and you can download it for free from GitHub or through the package managers listed above.

Do I need to know how to code to use it?

Using it directly takes some familiarity with the command line, even though it includes a web interface. If you'd rather avoid the terminal, you can pick apps with a graphical interface that work with the same models.

What computer do I need?

It depends on the model: small ones run on a recent laptop, large ones need a lot of memory and a good graphics card. llama.cpp works on Windows, Mac and Linux and supports cards from several manufacturers.

Official sources

This article is written with the help of artificial intelligence models, based on the official sources listed, and is reviewed before publication.

Condividi questo articolo sui tuoi socialShare this article on your social networksComparte este artículo en tus redes socialesTeile diesen Artikel in deinen sozialen Netzwerken

No products were found matching your selection.

Iscriviti alla newsletter di You vs AISubscribe to the You vs AI newsletterSuscríbete al boletín de You vs AIAbonniere den Newsletter von You vs AI