Free AI always comes with fine print.
I review AI tools for a living, so I’ve developed a reflex: whenever a tech giant announces something “for everyone,” I go looking for the asterisk. This week, Mark Zuckerberg gave me a fresh one to examine. Meta released Glimmer as an open-weight model — something anyone can actually download, inspect, and run — while pitching a broader message that AI should be accessible to everyone. Sounds great. But Meta also keeps Muse Spark, its proprietary model, firmly locked away. And that gap between the message and the product lineup is where things get interesting.
What “open” actually gets you
Let me give credit where it’s due, because I’m not here to dunk on open weights. When a company releases a model with open weights, that’s a genuinely useful thing for people like me and probably like you. It means developers, researchers, and small teams can build on it without begging for API access or watching a usage meter tick up. From a toolkit perspective, open-weight models are the difference between owning a wrench and renting one by the minute.
So Glimmer being open-weight matters. It’s a real contribution, not vaporware or a press release dressed up as a product. If Meta had released nothing, nobody could complain. They released something people can actually use.
The model behind the curtain
But then there’s Muse Spark. Meta’s proprietary model stays exclusive, and that changes how I read the whole “for everyone” pitch. Because the pattern it suggests is one we’ve seen before across this industry: give away the good stuff, keep the best stuff.
Think about what that means in practice. If the open model is the one you share and the closed model is the one you protect, then “AI for everyone” really means “some AI for everyone.” The community gets a capable tool. Meta keeps the crown jewels. That’s not a scandal — it’s a business — but it’s also not universal access, and we shouldn’t pretend the two are the same thing.
As someone who tests tools honestly, I care less about mission statements and more about what actually lands in users’ hands. And what lands in users’ hands here is a two-tier system: open access to one model, no access to another.
Why the framing matters
You might ask why any of this matters if Glimmer is genuinely useful. Fair question. My answer: framing shapes expectations, and expectations shape the market.
When the CEO of one of the largest tech companies on the planet says AI should be accessible to everyone, that’s not just a philosophy — it’s positioning. It tells regulators, competitors, and users a story about who Meta is. If the reality is more selective than the story, that’s worth pointing out, because plenty of smaller companies with genuinely open approaches don’t get the same megaphone.
There’s also a practical concern for anyone building on these tools. If you architect your product around a company’s open offerings, you’re betting that the openness continues. A company that keeps its strongest model proprietary is signaling where its real priorities sit. That doesn’t mean you shouldn’t build on Glimmer — it means you should build with your eyes open.
My honest read
So does Zuckerberg really believe AI is for everyone? My take: he believes some AI should be for everyone, and that belief happens to align nicely with Meta’s strategic interests. Open-weight releases build goodwill, seed ecosystems, and pull developers into Meta’s orbit. Keeping Muse Spark closed protects whatever edge Meta thinks it has. Both moves make sense. Together, they just don’t add up to “for everyone.”
Here’s how I’d score it as a reviewer. Glimmer’s release: a genuine win for accessibility, no caveats needed. The “AI for everyone” messaging: partially true, and the partial part matters. Companies should get credit for what they open up and honest scrutiny for what they don’t.
My advice for anyone in the toolkit trenches is simple. Use the open tools — they’re real and they’re valuable. But judge companies by their full catalog, not their most generous release. When Meta opens up Muse Spark, or something like it, I’ll happily update this take. Until then, “for everyone” reads more like a slogan than a policy. And slogans, in my experience, don’t run on your hardware.
🕒 Published: