Meta wants credit for being “open” on AI without paying the full price of being open. And honestly, I get why they’re trying to thread that needle. I just don’t think we should pretend it’s some principled stance. It’s a business move dressed up as a philosophy.
Here’s what’s been shared publicly: Meta released a new AI model that developers can download and modify. But the company is keeping its most advanced AI tech closed. The reason is not mysterious. Meta wants to make money from the expensive AI race, and it’s feeling pressure from rivals like OpenAI and Anthropic. So it’s offering something that looks generous to the developer world while protecting the crown jewels.
I’m torn in a very specific way. I like the part where developers can actually touch the model, shape it, and build real stuff. That tends to create more experimentation, more weird ideas, more useful tools that don’t have to beg a single vendor for permission. But I don’t like the part where “open” becomes a marketing word that really means: “We’ll share what’s convenient and hide what’s valuable.”
Because that’s the game here. If Meta truly believed open models were the future, it would put the frontier models out there too. It isn’t doing that. It’s saying, in plain terms, “We’ll open the layer that helps adoption, but we’ll close the layer that keeps our edge.” That’s not evil. It’s just not the same thing as openness.
And the consequences are bigger than a branding argument. This shapes who gets power in the next phase of AI.
Imagine you’re a small startup building a customer support tool for niche businesses. You pick Meta’s downloadable model because it’s flexible and you can tweak it. Great. You ship faster. You don’t have to pay high fees to a closed provider. But then your competitors also pick it, because it’s the obvious free-ish base layer. Now everyone’s product starts to feel the same. Differentiation moves up the stack, and the base intelligence becomes a commodity… controlled by whichever company decides what “open enough” means.
Or imagine you’re a big company deciding where to invest. If Meta keeps its best models closed and positions them as the premium option, the safest bet for many teams will be to align with the closed frontier—because executives like predictable vendors. That pulls talent and money toward a few giants. The “open” release then becomes a feeder system: it trains developers, normalizes Meta’s ecosystem, and creates demand for the closed upgrade.
That’s the part people don’t say out loud: partial openness can be a funnel.
To be fair, there’s another side. Releasing powerful models widely can cause real harm. Bad actors can use them. People can fine-tune them for scams, harassment, or worse. And once something is downloadable, you can’t pull it back. So yes, there’s a safety argument for not releasing the most advanced version.
But I don’t fully buy that safety is the main driver here, because Meta is also openly talking about monetizing its AI investments and competing at the frontier. This looks like “we’ll be open where it helps growth, closed where it helps revenue.” Again: rational. But let’s call it what it is.
What worries me most is the middle outcome: a world where we get neither true openness nor healthy competition. We get curated openness that benefits the same handful of companies. Developers get a sandbox, while the real power stays locked behind paid doors and private deals. That’s not the worst outcome, but it’s not the hopeful one either.
There’s also a cultural cost. When a company says “download and modify” with one hand and “most advanced is proprietary” with the other, it trains everyone to accept a new normal: openness as a feature tier. That makes it harder for anyone—another company, a lab, a government—to argue for genuine public access later, because the public will already be used to the idea that the best models are private property.
Still, I don’t want to pretend the alternative is simple. If Meta open-sourced its most advanced model tomorrow, a lot of people would cheer, and a lot of people would panic. Regulators would glare. Competitors would scramble. And Meta would be taking on risk that other frontier companies avoid by staying closed.
So we’re left with a messy reality: Meta is trying to play both sides because both sides have real benefits. Openness creates adoption and goodwill. Closed models create pricing power and control. The company wants both.
The question is whether developers and the public should reward that strategy with trust, or treat it like what it is: a calculated compromise that mainly protects Meta’s advantage—so what kind of “open” is actually worth supporting?