On Tuesday, Meta released Muse Spark 1.3, its first frontier model in a year, and the Hacker News thread did what Hacker News threads do: 590 points, 390 comments, and a collective lament for the death of open weights. The model is closed. The weights are not coming — at least not yet. Developers who built on Llama feel betrayed.

They’re mourning the wrong thing.

The number that should have dominated the discussion is not the closed weights. It’s the price: $0.10 per million input tokens, $0.20 per million output. Mark Zuckerberg called it “almost too cheap to meter.” For once, the marketing is not an exaggeration. Muse Spark 1.3 lands at #6 of 636 models on the Artificial Analysis Intelligence Index — and it’s priced like a commodity.

The Open-Weights Thesis Just Got Falsified

The open-weights movement has operated on a single, unexamined premise for three years: openness is the mechanism of democratization. If you want AI to be accessible, you release the weights. Meta was the champion of this. Llama was the gift that kept giving — the foundation on which thousands of startups, researchers, and hobbyists built.

Muse Spark 1.3 falsifies the premise. Meta just made frontier AI cheaper than the open-weights ecosystem can match. You cannot fine-tune a Llama derivative to beat Muse Spark 1.3 at $0.10 per million input tokens. The economics don’t work. The compute alone costs more than the API.

The developers on Hacker News are angry about the closed weights. But the weights were never the point. The point was always access. And Meta just found a cheaper way to provide access: sell the inference at cost, keep the weights.

This is the razor-and-blades inversion nobody saw coming. Gillette gives away the razor and sells the blades. Meta used to give away the razor — the weights — and let you make your own blades. Now Meta is selling the razor at cost and keeping the blades. The developers who built businesses on free Llama weights are the ones who lose. They’re now competing with a subsidized API they cannot match.

The Pivot Was Organizational, Not Ideological

The backstory matters here. Meta’s open-weights strategy was a product of the old AI org — the one that shipped Llama 4 with training data contaminated by benchmark answers, a scandal that forced a reorganization. In June 2025, Meta spent $14.3 billion for a 49 percent stake in Scale AI and brought in Alexandr Wang as chief AI officer. The new org has different incentives.

Scale AI is a data-labeling company. Its business model is not giving things away. The proprietary pivot is not a philosophical betrayal of the open-source movement; it’s what happens when you rebuild your AI lab around a company that sells access to data, not weights.

One engineer who worked on Llama derivatives put it to me in a Slack DM this week: “We didn’t lose the weights. We lost the subsidy.” That’s the sentence the Hacker News thread should have been arguing about.

The Third Path

The open-weights debate has always been framed as a binary: open or closed, commons or enclosure, democratization or capture. Meta just walked through a third door.

Muse Spark 1.3 is closed — and cheaper than the open alternatives. The concentration of power in AI is happening through pricing and distribution, not through weights. The “AI safety through openness” crowd, which argued for years that closed models are dangerous because they concentrate power, now has to explain why Meta’s closed model is the cheapest frontier option on the market.

The competitors have a problem too. Anthropic and OpenAI charge premium prices for models that Meta is now selling at commodity rates. If Muse Spark 1.3 is “almost too cheap to meter,” what does that do to the unit economics of every other frontier lab? Zuckerberg has already teased “Muse Spark open weights releases coming soon” — which, if the pricing holds, would make the open-weights debate entirely moot. Why fight over weights when the API is cheaper than the fine-tune?

The open-weights mourners are right about one thing: something did die on Tuesday. But it wasn’t the commons. It was the assumption that openness was the only path to democratization. Meta found a cheaper one. The question now is whether anyone else can afford to follow.

Sources