llama.cpp reaches version 1 after 3.5 years

The team behind llama.cpp released version 1.0.0 on the 13th of November 2026, Hugging Face said. The tool is free software that runs AI models on people's own computers.

It is the first stable release in the project's 3.5 years, the team wrote. Many popular local AI apps are built on it.

What changes

From now on, any change that breaks how other apps connect will raise the first number, the team said. That gives app makers clear warning before an update.

The ggml library underneath, which does the maths, also reached 1.0.0.

Wide reach

llama.cpp supports more than 150 model designs, the team said. It ships for Linux, Mac, Windows, Android and iOS.

Nvidia, AMD, Qualcomm, Arm, IBM and Intel give hardware for testing, the team said.

Why it matters

Running AI on your own computer keeps data at home and works offline. The team said local AI and privacy are becoming more important in daily life.

Below we break down the timeline

The project took 3.5 years, or 42 months, to reach a stable label, by our arithmetic. It now lists 8 ways to install it, from Docker to Homebrew.

What is confirmed and what is not

The release is on the project's GitHub page. Speed changes were not part of the announcement.

Key facts

llama.cpp version 1.0.0 came out on the 13th of November 2026.

It is the project's first stable release.

It supports more than 150 model designs.

Breaking changes will now raise the first version number.