How Uprouter Verifies Data: Our Methodology, in Plain Language

Data is only useful if you can trust it — and in a space full of marketing numbers, "trust" is doing a lot of work. This is our methodology in plain language: where our data comes from, how we normalize it, what we mark as verified versus estimated, and the hard rule that keeps us honest. Read it and you'll know exactly how much weight to put on any number you see on the site.

The guiding rule: data-true, never pay-to-rank

The single most important thing about Uprouter: inclusion is never sold, and no listing position can be bought. Entries are ordered by data — free-credit value, price, risk rating, community score — not by sponsorship. Affiliate relationships exist on some entries and are disclosed, but they never affect the data, the scores, or the sort order.

That rule is what lets the rest of this page be meaningful. If rankings were purchasable, the "verification" underneath them would be theater.

Where the data comes from

Our directory is built from a few sources, and we're explicit about which is which:

• Official sources. Provider pricing pages, docs, and status pages are the primary reference for prices, model specs, and terms. We record the source URL on each entry so you can check it yourself.
• Normalization by us. We convert everything into a comparable shape — USD per 1M input/output tokens, a quantified free-tier value, a risk tier — so you're comparing facts, not marketing copy.
• Community input. Signed-in users can upvote, write evidence-backed reviews, and propose corrections. These feed the signals (votes, review counts) but never overwrite the objective data fields (price, risk, status).
• Our own probes. We run best-effort status and latency checks against endpoints. These are measurements, and they're always labeled as such — see what "up" really means.

Verified vs. estimated: we don't hide the distinction

This is where a lot of directories quietly lose trust, so we're blunt about it:

• Verified — a figure taken directly from an official, dated source. The entry shows when it was last verified.
• Estimated — a figure that required an assumption on our part. The most common case is free-tier value: a free allowance is measured in tokens or requests, so we convert it to dollars using the provider's own pay-as-you-go rates for the models a developer would realistically use. When we do that, the number carries an explicit "est." flag.
• Unknown / not quantified — when we can't quantify something honestly, we say so. A free tier that doesn't convert cleanly to dollars is shown as "not quantified," not as a made-up number.

The principle: a flagged estimate and an honest "unknown" are both better than a confident wrong number. If a number matters to you and it's marked estimated, check the source link — it's there for exactly that reason.

How risk is rated

The risk tier (low / medium / high / unrated) is an editorial assessment, distinct from a live measurement. It weighs evidence like: how long the operator has run, how transparent its terms are, its reliability record, and clarity of data handling. The reasoning is shown on each entry, not buried.

Keep the distinction straight: live status is a measurement (it can flip any second), risk is an assessment (it changes less often and reflects the bigger picture). Both are starting points for your own judgment, not guarantees — and the risk rating is never influenced by affiliate relationships.

What we deliberately don't do

• We don't fabricate model names, prices, or statistics. If we don't have a source, we don't print a number.
• We don't present probes as SLAs. Our status checks are best-effort from our vantage point — production decisions should use the provider's own status page and your own monitoring.
• We don't endorse. Listing a provider is not an endorsement; everything on the site is informational, and we say so on every page.

Why the methodology matters to you

When you use this site to compare routers or providers, the methodology tells you how to read the results: trust the verified, dated, source-linked numbers; treat estimates as order-of-magnitude; rely on the honest "unknowns" as a sign of care rather than a gap. And remember the frame — you make the final call on any provider's terms before you spend money with them.

You can propose a correction on any entry if you spot something stale or wrong — community input is part of how the data stays current. The full policy lives on the about & methodology page, and corrections can be sent via contact.

The bottom line

Uprouter is data-true by rule: official sources first, normalization second, estimates and unknowns labeled honestly, and inclusion that is never for sale. That's the whole methodology. It's not a promise that every number is perfect — it's a promise that when a number isn't perfect, you'll know, and you'll be able to check the source yourself.

Related reading

• How We Check AI API Provider Reliability
• Understanding AI API Pricing
• How to Choose an AI API Router in 2026

FAQ

How does Uprouter decide the order of entries? By data only — free-credit value, price, risk rating, or community score, depending on the view. Inclusion is never sold and no position can be bought; affiliate relationships never affect sorting.

What does the "est." flag on a free-tier value mean? It means we converted a token/request-based allowance into dollars using the provider's own pay-as-you-go rates, which involves an assumption. It's an order-of-magnitude figure, not a quote — check the source link for the exact terms.

What's the difference between live status and risk tier? Live status is a recent measurement from our probes (it can change any second). Risk tier is a slower, evidence-based editorial judgment about the operator and its terms. Neither is a guarantee; both are inputs to your own decision.

How can I correct data I think is wrong? Use the edit/correction option on any entry to propose a change with evidence; admins review proposals and apply verified corrections. You can also reach the editors via the contact page.