About Attic Standard
Buyers of AI inference cannot answer the question that buyers in every other infrastructure market take for granted: is this price normal. Vendors publish in units of their own choosing, per million tokens here, per image or per second of video there, some with a cached tier and some without. Rates move week to week, and the moves are announced by whoever benefits from them. There is no neutral series to check any of it against.
What follows is what always follows. Buyers overpay because they have nothing to compare against. Finance teams cannot forecast a line item that changes without warning. And the vendors who move first end up setting the reference price for everyone else, which is a position no competitor can argue with and no customer can audit.
Attic Standard publishes an independent price benchmark for AI inference, constructed the way financial indexes are constructed. Every week we verify the published rates of every vendor we track, normalize them into comparable units, and chain them into indexes that separate genuine repricing from changes in what happens to be tracked that week.
Nothing is estimated. Nothing is inferred from a model's behaviour or guessed at from a scrape. If a vendor does not publish a rate, it is not in the index, and every value that is in the index can be traced back to the published page it came from and the date it was read.
The result is a series you can argue with, because the construction is public and the inputs are verifiable. That is the whole product. Everything else we build sits on top of it.
No vendor sponsors an index. No vendor pays for inclusion, and none can pay to be left out. No commercial relationship influences methodology, and no vendor sees a number before it is published.
This is a constraint on the business rather than a line of marketing. It rules out the revenue model most price-comparison sites run on, which is to charge the companies being compared. We took that trade deliberately, because a benchmark that can be bought is not a benchmark, and the moment our numbers become negotiable they are worth nothing to the people who rely on them.
This page is the argument. The methodology page is the evidence: inclusion criteria, exclusions, normalization rules, the chained matched-model calculation, and the glossary that fixes what each term means. The homepage carries the indexes themselves, live, with current coverage and weekly movement.