Model prices collapsed again. Your strategy shouldn’t care.

This month brought another round of frontier-model price cuts — the steepest reported at 80% (industry roundup). If your AI strategy changes every time the per-token price does, you have a bet on the wrong layer.

The latest drop

Every few months the pattern repeats: a new model ships, benchmarks jump, and the price of last quarter’s state-of-the-art falls off a cliff. Capability that cost dollars per million tokens in 2024 now costs cents. Whatever number you are paying today, plan for it to be embarrassing by spring. This is not a market anomaly — it is the market working.

Why prices keep falling

Because intelligence-by-the-token is a commodity with at least four credible suppliers, and commodities compete on price. Each provider’s frontier advantage lasts a quarter or two before the others match it. For a buyer, that is wonderful — and it carries a hard strategic implication: anything you build whose value depends on which model you chose is a depreciating asset.

Where the moat actually is

What does not commoditize: your workflow encoded into a system — the integrations into your existing software, the evaluation set built from your historical cases, the compliance rails a regulator will accept, the trust your team places in a tool that has run beside them for months. Model prices falling makes that layer more valuable, not less: every price cut lowers the operating cost of a well-built agent while the switching cost of the workflow layer stays put. This is the core of our thesis: models commoditize, workflows compound.

If you are buying AI

Three rules. Demand model independence — any agent you commission should be able to swap engines as prices fall, so the savings flow to you. Pay for the workflow, not the wrapper — a thin UI over an API is worth exactly the API minus the markup. Judge vendors on evals, not demos — the demo shows the model, which everyone rents; the evaluation set shows the workflow, which is the thing you are actually paying for. That is how we build, and it is why falling prices are our clients’ favorite news.