FactQuire

A facts wire for the model economy: why we exist

2026-07-04

A facts wire for the model economy: why we exist

Every week, model prices change, context limits move, and models are quietly deprecated. And every week, thousands of routers, gateways, cost dashboards, and agent frameworks make real spending decisions using metadata that nobody verified.

The de facto standard today is a community-maintained JSON file — a heroic volunteer effort that most of the ecosystem consumes as if it were ground truth. It has no verification step. Most entries carry no evidence of where a number came from. When it lags or errs, the error propagates silently into everything built on top of it.

That is a strange foundation for an economy this large. So we built the boring alternative.

What this is

A machine-readable feed of facts about commercial LLM APIs — prices, context windows, output limits, lifecycle status, deprecation dates — with a discipline borrowed from wire services, not from web scraping:

What we found on day one

As a first exercise, we compared the most widely consumed community datasets against primary sources, entry by entry, and adjudicated every mismatch — the results are on the audit page, including any errors of our own, listed first. We intend to run this audit continuously and to contribute confirmed fixes upstream. The community datasets are valuable; our job is to make the numbers in this ecosystem checkable.

How to use it

The full dataset is one file: feed.json. Each model has a stable permalink. If you maintain a gateway, router, or cost tool, you can consume the feed directly or simply diff it against your current source — the evidence quotes make disagreements fast to settle. The citation guide covers the rest.

Facts for people, and for the software that increasingly acts on their behalf. That's the whole business.