The most hyped AI model of the year already has open versions you can download and run yourself.
For the past week my feed has been one long celebration of a new AI model called Jev. Fastest launch of the year. A whole new category. Forty million dollars in fresh funding. Almost nobody is mentioning the part I keep staring at: you can already download an open version and run it yourself.
First, what it is, quickly. Jev makes fast, cheap decisions for software, the small yes-or-no, which-bucket, is-this-safe calls that software makes a million times a day. Useful, and real demand. But here is what stopped me. More than a year ago, an independent researcher had already published the same idea in the open: a small model that answers with a probability instead of a paragraph, the model, the data, and the method, free for anyone to use. It got a handful of comments and no funding.
Then Jev launched to a wall of attention, and the open world answered within days. Dozens of alternatives appeared. One of them, Laya, comes from that same overlooked researcher. Openly licensed, free to use and change, small enough to run on a laptop, two commands to get it running.
Why so fast? Because the hard part was never a secret. These models sit on open foundations anyone can download, the same free base models the whole industry trains on. The paid product is a layer of polish on a public floor. There was no vault. There was a head start.
Now the honest part. Today Laya is far narrower than Jev. It reads a few hundred words of input where Jev reads tens of thousands, so for now it suits short, high-volume decisions, not long documents. Out of the box it is rougher, and matching the paid product on your own work takes tuning it on your own data.
But that ceiling comes from the small model it was built on, not from a lock someone else holds. Because it is open, it can be retrained, extended, rebuilt on a bigger base. You cannot do any of that to a closed product you only rent. The one thing standing between you and that independence is data you already own.
For a bank, a hospital, or a public agency that cannot legally hand its data to an outside server, this decides whether they get to use the technology at all. And the open version they need almost always exists already, usually within days of the headline.
It is worth turning that same question on your own stack. How much of what you rent from a single vendor, one that can raise the price, change the terms, or disappear tomorrow, could you already be running yourself? For most of us the honest answer is more than we think, and the only reason we do not know is that we never looked.
