llama.cpp

v0.8.0Open sourceReleased 28 Oct 2026

The C/C++ engine behind most local LLM apps, with a CLI, server and web UI. By ggml-org

Windows
macOS
Linux

Official sourceNo bundles

By the ToolsBharat teamUpdated 28 Oct 2026How we check

Quick answer

llama.cpp is open source software for Windows, macOS and Linux from ggml-org. The C/C++ engine behind most local LLM apps, with a CLI, server and web UI. Download it only from github.com. The latest version we track is 0.8.0 (28 Oct 2026). ToolsBharat links straight to the publisher's own download page, with no bundled offers.

Latest version:: 0.8.0 (28 Oct 2026)
Publisher:: ggml-org
Official site:: github.com
Licence:: Open source — MIT
Requirements:: Windows, macOS, Linux

Our verdict

llama.cpp is the inference engine that many desktop AI apps are built on, and using it directly gets you the newest model support first. Prebuilt zips cover Windows (CPU, CUDA, Vulkan and more), macOS and Linux, and the bundled server offers an OpenAI-compatible API plus a simple web chat. It is a command-line tool at heart, so beginners may prefer a GUI app built on top of it.

Screenshots

llama.cpp web UI chat screen — Official GitHub repository

llama.cpp running a model in the terminal — Official GitHub repository

Key features

Runs GGUF models on CPU, GPU or a mix of both
Backends include CUDA, Vulkan, Metal, HIP (AMD), SYCL (Intel) and more
OpenAI-compatible API server with a built-in web UI
Quantisation from 1.5-bit to 8-bit to fit models into less memory
Can pull models straight from Hugging Face

Pros

Pro: Very fast-moving, with new model architectures added quickly
Pro: Separate builds for Windows on x64 and ARM64
Pro: No Python environment needed for the prebuilt binaries

Cons

Con: Command-line first; no installer wizard
Con: Many build variants can confuse first-time users
Con: New builds appear several times a day

What's new in 0.8.0

See what changed in 0.8.0 in the official release notes

Version history

Version 0.8.0Latest
28 Oct 2026
SHA-256
The developer does not publish one
Source
Release notes

Official downloads by platform

For Windows, use the Download button. Other platforms go straight to the publisher's own page or official store.

Requirements

OS: Windows, macOS, Linux
Architecture: 64-bit, ARM64
Also needs: GPU optional; pick the CUDA, Vulkan or CPU build that matches your hardware

Licence

Type: Open source
Licence: MIT
Publisher: github.com

Where these facts come from

Checked 30 Oct 2026

FAQ

Which file should I download on Windows?

The releases page lists separate zips per backend. The CPU build works on any PC; NVIDIA users can pick a CUDA build (with its matching cudart package), and Vulkan builds cover many AMD and Intel GPUs. Numbered bNNNN builds are published several times a day, while versioned releases such as v0.8.0 summarise the changes.

User reviews

User reviewsयूज़र रिव्यू

No reviews yet

Used llama.cpp? Reviews appear after moderation; we never import or invent ratings.

Alternatives