Your Model Is a Situationship

But it’s time to get serious.

You know the type. Dazzling at first, attentive, captivating, there at any hour, exactly when you need it. And then one day it goes silent. You’re sure you did something wrong. But you did nothing, and you were never going to get an explanation anyway.

This is exactly what happened on a Friday night in June — to thousands of companies at once. The one who went quiet was a dazzlingly powerful new model, forced into silence by a government order days after it shipped. The companies had done nothing wrong. There was nothing they could do. A kill switch is not a plan, but in June someone threw it. It texted back weeks later, but only because a government let it, and never once because you asked.

You see, you’re in a situationship. The closed model you rent is brilliant, attentive, and seeing thousands of other people at once. You keep going back for a reason: It can give do anything you ask. It can be fun while you’re still figuring out what you want. Until one day you want more, and you ask where this is going: “What are we doing here?” That is the one question a rented model cannot answer. The companies that were ghosted in June were not wrong to love a closed model. Their mistake was having nowhere else to turn.

There is always a friend who says not to overthink it. Open, closed, who cares? Just ask whether the model is good. It sounds reasonable. But take that advice and you give up the upside, because how you hold a model decides whether you date it forever or build something that is yours. And that is where the whole opportunity lives.

A model is not something you grade once and take home. It is a relationship, and relationships come in every level of commitment. The one you rent is the situationship you already know. You can take the good habits from one into the next, the lessons without the baggage. You can bring one home for good, a copy that moves in and cannot be put out on the street, even if you never quite learn where it came from. You can make it yours, reshaping it until it fits your life and no one else’s. You can build your own from scratch, with nothing about it hidden from you. Or you can build one with everyone at once, the way all of computing was created, a thing with a thousand hands on it and no one person who can carry it off.

These are not better and worse. They are different relationships for different chapters, and the only real mistake is not knowing which chapter you are in. You date around while you’re still figuring out what you want. You define the relationship when you are ready to build a life on it, because you do not put a situationship into production.

No One Can Read a Model’s Mind

Start with the model in front of you, because the first thing a real choice gives you is the power to check instead of blindly trust. You are told the closed one is safe by the company that sells it, and that is a promise, not proof. The cleverest case for closed starts true. No one can read a model’s mind, not even the lab that built it, so seeing inside buys you nothing. That part is right. The rest does not follow, because reading its mind was never the point, and no one can do that anyway. What open hands you is everything short of it. You can pin a model so it cannot change overnight, run it where nothing you say leaves the room, and test it against your own questions. That is the difference between being promised you are safe and genuinely knowing it.

One honest cost comes with it, and it is the best card the other side holds: A model set loose cannot be recalled, so for the rare capability that could help someone build a weapon, the wall is the right call and open should wait. That line is real, and it is narrow. It does not touch the everyday model that drafts your email and writes your code, which is nearly all of what is at stake. The fight was never open against closed everywhere. It is verify against trust, and open is the side that lets you verify.

But you can only check a model you are holding, which changes the question from what you can see to what you should keep. The most impressive model is the life of the party; the model that matters is the one that’s still there in the morning, the one that answers the thousand boring questions, lives inside your apps, and does the work long after it stops being exciting. The frontier is where the sparks are. The base — the one you run for pennies and keep for good — is about real life.

You can see where the builders are drifting, not from what they say but from where they spend. On OpenRouter, the big American closed models have slid from about 70% of the volume a year ago to roughly a third, while cheap open ones now carry most of it, many for a tenth or a twentieth of the price and landing within a point of the famous names on the hard coding tests. No single model owns even a fifth of the traffic. This is not the whole market. Closed still owns the enterprise deals and the apps on your phone. But the people who feel the cost are quietly going home with the one they can build a life with.

History Doesn’t Have to Repeat Itself

This is the oldest story in computing, and you already know how it ends. In the late 90s. Linux lost almost every argument about being better, and the companies selling the closed systems said — nearly word for word what you hear now — that the free thing was not really better, that you still needed real machines and someone to make them fast. They were right, for a while. Then they were trivia. Linux did not win by being the best. It became the soil the whole internet grew out of, because anyone could build on it without asking. (Try growing an internet on rented land.) The open web ran the same bargain. No one owned the roads, so a billion things got built on them that no owner would have ever approved. A closed foundation, however brilliant, can only grow the tidy monocrop its owner planted. An open one grows a jungle. And the jungle, not the soil, is where the value lives. The opportunity was never in owning the ground. It was in everything that grows once the ground is free for anyone to build on.

So let the closed labs keep the frontier and charge a fortune for it while it leads. That premium funds the next run, the way every chip generation has paid for the one after it. When open catches up and the selling window shrinks, the run gets paid for less by access and more by the strategic and national money that wants the lead for its own sake. Either way, last year’s frontier sinking into this year’s open base is the system working as designed, not a leak to plug. The model is a melting asset. The money is in what gets built on top once models are cheap. What has to stay open is the ground we all stand on, because the minute it becomes one company’s land, the jungle stops growing.

Which is the whole game, because the ground we all build on is about to be the ground we all think on. A model is becoming a cultural technology, like the printing press and the dictionary: the thing between a person and what they write. (Not to mention defining norms.) So the real question stops being how clever the model is and becomes who gets to set its defaults — what it calls true and what it refuses. A world where a few companies decide that is a much smaller one.

A model does not have to be cruel in order to flatten you. It just has to be the only one — one that doesn’t quite speak your language. Look at Linux: One kernel became a watch and a phone and a car and every supercomputer on the planet, bent into 10,000 shapes by people who each wanted a different machine. Open does not give you a better default, but, more importantly, it lets you set your own. The honest part — the part the cheerleaders skip — is that setting your own is work. Somebody still has to build the model that speaks a language the frontier skipped, and open makes that possible, not easy. But possible is everything, because the only other setting is forbidden.

Most of the cheap open models eating the market today come from a single country. The enemy was never another country’s flag. It was monoculture itself — ours and closed or theirs and open — and the fix does not change. It requires many hands, some of which have to be yours, because the model that your language or your field needs most is one the giants will never build, and open is the only place you don’t have to wait for them. So build it. Even then, it will not hold still. The open web gave us a thousand voices, and then a handful of platforms herded most of them into a few feeds. The pendulum swings back toward the middle the second something is built on top. What an open foundation buys you is a pendulum still free to swing. Diversity is never won and shelved. It is won, and lost a little, and won again, by whoever keeps showing up to build.

Welcome to the Defining Moment

And none of it is settled yet. We are at the moment the foundation gets poured, the concrete still wet and able to take the shape of our hand. That is simultaneously the hope and the deadline. We must keep the foundation open.

So stop asking which model is best and shiniest and start asking the questions of someone who is done drifting: Where is this going, and can I walk away if I need to? The answer is yours to choose, on every floor of what you build. The future to want is the grown-up one, where you love the frontier for the wonders only it can make, you build your life on a base you own, and can stay free, leaving only on your terms. Weigh the argument, not the source. (That goes for the one you’re reading, too.) The next time someone calls open source a distraction, remember that this is what they said about Linux.

Download the weights while the downloading is good. Run a model on a machine that is yours. Never build your life on something you cannot walk away from. The concrete will still be wet for one more season, and the people who want to set their foundation in it are counting on you keeping your hands out of it. The worst way this ends is the slow fade, where nobody chose anything — everybody just kept renting because renting was so easy — and one morning the concrete was set. You were not born to rent the thing you think with. The other ending is not a forecast. It is a to-do list. Go build it.