Price the router separately from the thing it routes to

The same clip, from the same model, cost $1.76 bought through an aggregator's credits and about $0.40 bought direct.

There is a product shape that has quietly taken over the AI video category, and it is worth naming before you buy one: the router.

A router does not make the thing. It sells you one balance, one login and one bill, and spends that balance across a shelf of other companies' models. The pitch is obvious and genuinely good — you get to try fifteen models without holding fifteen subscriptions, and you stop maintaining a wallet of half-used credits at five vendors.

What is easy to miss is that a router has to make its margin somewhere, and it makes it on the exchange rate between your balance and the underlying model.

So I priced one identical job both ways.

Same model, same clip. Through the aggregator's credits, at the rate a mid-tier annual plan pays, that clip worked out at $1.76. Bought directly, the same model's own hosted price for the same job is about $0.40.

That is not a scandal. It is a number, and it is the number the decision should turn on. Four-ish times is a lot to pay for routing if you have settled on one model and generate every day. It is nothing at all if you are still deciding, because the alternative — subscribing to four vendors to run a bake-off — costs far more than the markup on a handful of test clips.

The honest version of the advice is therefore not "aggregators are a rip-off". It is: the convenience has a price, that price is computable, and almost nobody computes it.

Here is the exercise, and it takes about ten minutes.

1. Pick one job you actually run — a specific model, a specific length, a specific resolution. Not a hypothetical.
2. Work out what that job costs inside the aggregator. You will usually have to convert credits to money via your plan's rate, which is the step the pricing page does not do for you.
3. Find the same model's direct price for the same job.
4. Divide. That ratio is what you are paying for routing.
5. Multiply by how many of those jobs you run a month. That is your annual convenience bill, and it is the only number that should decide whether you stay.

Two cautions from doing this properly, both of which cost me time.

Compare the same model, not the same family. Model names in this category are a minefield of versions and speed variants, and the price gap between a flagship and its "fast" sibling is often larger than the gap between two vendors. A comparison that quietly swaps versions produces whatever answer you were hoping for.

Check whether the cheap direct price is an introductory one. At least one of the cheapest subscriptions in this category is a first-term promo that renews materially higher. A markup measured against a promotional rate is not a markup, it is a countdown.

One thing I want to be straight about, because it shapes what this comparison can tell you. We generation-tested the alternatives; we did not generation-test the aggregator itself. That means I can compare what a clip costs through each route, and I cannot tell you whether the clip that comes out is the same on both sides. Routing usually means the same upstream model, but "usually" is doing real work in that sentence, and I would rather flag it than let the price comparison imply a quality comparison I did not run.

The full roundup — the seven alternatives, which ones we actually generated with, where each one wins, and the two that are genuinely different decisions rather than variations on the same one — is in our Higgsfield alternatives comparison.

The general habit, well beyond video: whenever a product's job is to be a front end for other people's products, price one identical unit of work on both sides before you commit. The convenience is real. It should still be a purchase you make on purpose.