Z.ai's New Free AI Handles Text, Images, and Video — at a Fraction of the Cost

A new open-source AI model called GLM-5.3-Flash just dropped, and it's a big deal for anyone who uses AI tools. Made by Chinese lab Z.ai, this model can handle text, images, and video in a single system. That's useful for everything from summarizing meetings with slides to describing what's happening in a video clip. The company claims it performs nearly as well as some of the priciest AI models out there, but at a fraction of the operating cost.

What makes this model special is its clever engineering. It uses a mix of two attention methods to process long documents without needing huge amounts of computing power. For example, it can handle over a million tokens of context — think of reading an entire novel series in one go. It also uses a trick called "sparse attention" to focus only on the most relevant parts of a long text, which saves money and speeds things up. The result: AI app builders can offer powerful features without charging you an arm and a leg.

Another plus is that the model's code and weights are free to use under a permissive license. That means anyone — from hobbyists to startups — can download it and build their own AI tools, rather than relying on a big tech company's expensive API. It's a bit like getting a professional-grade engine for free and being allowed to install it in any car you want.

Of course, there are catches. The model is huge — about 331 gigabytes in its standard version — so you'll need serious computer hardware to run it yourself. And while it's close to top-tier performance, it's not quite the best in every category. Still, for people who value privacy, customization, or lower costs, this release is a promising step toward more open and affordable AI.