In mid-June, Z.ai released GLM-5.2 — an open-weight AI model that matches frontier coding benchmarks while costing roughly one-sixth what closed, premium models charge per token. Anyone can download the weights and run it on their own servers, for free. It’s not a discount. It’s a different kind of competition entirely.
For years, the pitch was simple: “we use the best model” was itself a competitive edge, because the best model was expensive and hard to access. That logic breaks the moment the best model becomes something anyone can download for free. If your neighbor has the same ingredients you do, “I have good ingredients” stops being something to brag about. The advantage has to live somewhere else.
That somewhere else is the kitchen, not the pantry — the layer that takes whatever model is available and actually makes it useful for a specific job. A company that has spent years building the pipes connecting AI to real workflows doesn’t lose that position just because a cheaper model showed up. If anything, more model competition raises the value of whoever’s doing the cooking, since no one wants to manage five different AI providers themselves. The same logic plays out one level up in the AI stack — Anthropic’s approach centers on workflows where getting the right answer matters more than speed or cost, which is exactly the kind of “kitchen” advantage that survives a cheaper model entering the market, as explored in an earlier look at AI export controls and platform investors.
The filter for any AI-adjacent investment: if the pitch is “we use a good model,” that’s a fading advantage. If the pitch is “we’re the layer that makes any model useful,” that’s an advantage that gets stronger, not weaker, as the models underneath keep getting cheaper.
A bad workman blames his tools.
Not advice. Just how I see it.