Free AI Tool llama.cpp Now Runs Faster on Your Phone

llama.cpp is a free, open-source project — think of it as the engine that lets ordinary computers and phones run AI chatbots locally, without sending anything to a company's servers. It's one of the most popular AI projects in the world, with roughly 129,000 GitHub stars (the equivalent of "likes"). Its latest build, b11147, includes a small but useful change: a new "kernel," which is a specialized set of instructions the computer uses to do AI math faster. This one targets OpenCL, software that lets programs borrow the power of graphics chips.

In practical terms, that means AI running on the graphics chip inside Snapdragon phones and Windows-on-ARM laptops — the new wave of thin, battery-friendly PCs. Graphics chips are very good at the kind of repetitive math AI needs, so using them well is the difference between a chatbot that stutters and one that answers quickly. The release also ships ready-to-install packages for Windows, Mac, Linux, Android and iOS, so developers don't have to build it themselves.

For most people, this changes nothing today. It's a pre-release developer version, aimed at tinkerers and app makers, and the speed gain is a modest tuning improvement rather than a breakthrough. If you use ChatGPT, Claude or Gemini through a website or app, you'll never touch this. There's also no new feature here — no smarter answers, no new abilities, just less waiting on certain devices.

The bigger story is the direction of travel. More and more, AI is moving off distant servers and onto the device in your pocket. That means privacy (your questions never leave your phone), savings (no monthly subscription), and offline access (it works on a plane). Each small update like this pushes that future a little closer. If local AI interests you, apps like LM Studio and Ollama make llama.cpp easy to try without any coding.