For a few years, arriving first has been a lasting advantage. The best model was scarce, expensive to achieve and difficult to imitate, and that scarcity translated into the price. Whoever arrived first got paid for it. A few days ago that premise was blown up with the silencer on.
Moonshot has presented these days Kimi K3a model with 2.8 billion parameters with the weights announced openly. It slips into second or third place in intelligence indices depending on who you ask, only behind Claude Fable 5 and GPT-5.6 Sol. And it leads the ranking frontend.
The dominant reading, expected in the middle of 2026, is geopolitics: “China catches up with the United States.” Well. It is the most obvious and the least interesting reading, if you ask me, because what is relevant is not the little flags but rather the distance between the closed border and its open replica has gone from the 6 to 9 months that were debated to something closer to between 3 and 5. If anything. The premium for arriving first is no longer measured in years. Not in months either. Or less and less. It is already measured in weeks.
Capacity is already getting cheaper before it becomes profitable
The business model of selling access to the best model rests on an assumption: that capacity will remain scarce long enough to pay off what it cost to achieve it.
(…)
GIPHY App Key not set. Please check settings