llama.cppb12627
Run large language models locally with C/C++ inference
136k
Stars
16.9k
Total downloads
10
Versions
6h ago
Page synced
Download
b12627Pre-release only2026-11-22 · 7h agoRecommended for most users
llama-b12627-bin-win-vulkan-x64.zip
37.1 MBx64408 downloadsSHA256 4e89c02cff…This is the right pick for most people, including when you are unsure about x64 vs arm64.
Windows — all files
cudart-llama-bin-win-cuda-12.4-x64.zip
373 MBx64239 downloadsSHA256 8c79a9b226…cudart-llama-bin-win-cuda-13.5-arm64.zip
266 MBarm6421 downloadsSHA256 f89e7b2457…cudart-llama-bin-win-cuda-13.5-x64.zip
407 MBx64216 downloadsSHA256 c2bf4cc202…llama-b12627-bin-win-cpu-arm64.zip
12.5 MBarm6415 downloadsSHA256 2ef3617106…llama-b12627-bin-win-cpu-x64.zip
23.9 MBx64290 downloadsSHA256 323bad725d…llama-b12627-bin-win-cuda-12.4-x64.zip
341 MBx64344 downloadsSHA256 e511b880ee…llama-b12627-bin-win-cuda-13.5-arm64.zip
203 MBarm6413 downloadsSHA256 825f727089…llama-b12627-bin-win-cuda-13.5-x64.zip
215 MBx64350 downloadsSHA256 8f42155f8b…llama-b12627-bin-win-opencl-adreno-arm64.zip
13.5 MBarm6410 downloadsSHA256 f436fedee7…llama-b12627-bin-win-openvino-2026.4.1-x64.zip
100 MBx647 downloadsSHA256 269d085cf5…llama-b12627-bin-win-rocm-10.0-x64.zip
271 MBx6440 downloadsSHA256 f4871a170f…llama-b12627-bin-win-sycl-x64.zip
165 MBx6421 downloadsSHA256 98b2e1920d…llama-b12627-bin-win-vulkan-arm64.zip
25.3 MBarm6411 downloadsSHA256 ed74dd673a…About this project
llama.cpp is a large language model inference tool written in C/C++. Its job is to run large models on your own computer, so instead of relying on a cloud service for everything, you can chat, write and experiment with a model locally, with your data staying on your own machine.
It is a great fit for people who want to try large models locally, whether for privacy, for offline use, or out of curiosity about how models run under the hood. Download the build for your system from this page, prepare a model file following the project's instructions, and you can start running models on your own hardware.
Key features
- Runs large language models on your own computer
- Written in C/C++ with a focus on inference
- No cloud service needed to run models
- Your conversations stay on your device
- Available across platforms
Use cases & who it's for
- You want to run a large language model offline on your own PC
- You care about privacy and prefer not to send conversations to the cloud
- You are learning how local LLM inference works
Maintenance
Actively maintained
Last release 7h ago
License
MIT
Permissive license
Allows:Commercial use、Modification、Distribution
Requires:Keep the copyright notice
Facts
- Language
- C++
- Forks
- 25.9k
- Latest release
- 2026-11-22
- Created
- 2023-03-10
FAQ
Is llama.cpp free?
Yes. llama.cpp is free and open source under the MIT license, and you can download and use it for free.
Which platforms does llama.cpp support?
llama.cpp supports Windows, macOS, Linux and other platforms. Choose the file that matches your system.
How do I download and use llama.cpp from this site?
Pick the file for your system from the version list on this page and download it. It is a command-line inference tool, so you will also need to prepare a model file following the project's instructions, then run it in your terminal.
Do I need an internet connection to run models with llama.cpp?
Inference happens on your own machine. Once you have a model file ready, you can run it locally without depending on a cloud service.
Older versions 9 versions
Related software
An AI assistant that really gets things done for you
The open source coding agent.
Run large language models like DeepSeek and Qwen on your own computer
All-in-one desktop assistant for Claude Code, Codex, OpenCode and more
