This Free AI Writer Just Got Faster on Your Own PC

Most people know AI assistants like ChatGPT, but those run on huge company servers. GLM-4.5-Air is different: it's a powerful AI writing model that you can download and run entirely on your own computer. Until now, it worked but felt a bit slow. A new update to the free software llama.cpp (the engine that runs such models) adds a feature called MTP, which roughly lets the AI predict several words at once. The result is a noticeable speed boost, as if your local AI suddenly got a caffeine shot.

Why does this matter to you? Because running AI locally means your writing stays private, there's no monthly subscription, and you don't need an internet connection. The model is also surprisingly efficient. Although it's technically 106 billion parameters (the knobs that give it intelligence), it only activates 12 billion per task—like a huge restaurant that only lights up the kitchen when you order. So a computer with lots of memory but average processing power, such as AMD's Strix Halo or NVIDIA's DGX Spark, can handle it comfortably. Even a single RTX 3090 graphics card works well.

Fans of the model are mostly people who enjoy creative writing, storytelling, or roleplaying with AI characters. The project leader notes it remains "great for creative writing" and points to dozens of free community-made versions tuned specifically for fiction and dialogue. There's even a link to download a small extra file if your copy doesn't include the new speed feature. Full installation is still a bit technical, but for those willing to try, the payoff is a private, fast, and free writing companion that feels far more personal than a web chatbot.