Blog

We graded the OpenRouter listicles

We read the eight pages that rank for OpenRouter alternatives and checked every fee claim against the vendors' own live pricing pages on a single day. Three were correct, two were wrong, and three never stated a number at all. Including the one we got wrong ourselves.

By the Unbiased Team · published August 25, 2026 · every figure verified against its primary source on August 25, 2026

We audited eight articles ranking for OpenRouter alternative queries against OpenRouter's own pricing documentation on 2026-08-25. Three stated the 5.5% credit fee correctly, two carried factual errors (a stale BYOK request threshold and a false claim that OpenRouter takes no crypto payments), and three contained no fee figures at all.

Why we checked

OpenRouter changed its bring-your-own-key terms. The old structure charged 5% on BYOK usage above a threshold measured in requests: the first million each month were free. The current structure, in OpenRouter's own documentation, is a $25,000 monthly allowance measured in list-price inference cost, with 5% above it, and a $200,000 allowance on enterprise plans.

That is not a cosmetic change. A request-count allowance and a dollar allowance behave completely differently: a million cheap classification calls and a million long agent turns are the same under the old rule and orders of magnitude apart under the new one. Anyone budgeting from the old number is budgeting from fiction.

So we went looking for who else still had it wrong. The honest answer turned out to include us.

Audited 2026-08-25. Fee claims quoted from each page as published; verdicts assessed against OpenRouter's own documentation the same day. Dates are each page's own stated publish or update date.

We had it wrong for ten days

Our own pages carried the request-count threshold from publication on August 9 until we re-verified on August 18. Nine pages said it, including the flagship comparison. Nobody caught it externally; we caught it because we ran a scheduled fact re-check against the vendors' live pages, which is the only reason this post exists.

That is the whole argument for dated claims. Every number on this site carries a verification date, so when a vendor moves, the staleness is visible instead of invisible. We were wrong for ten days and can tell you exactly which ten.

What the audit found

The table above is the full result. The pattern worth naming is not that two pages were wrong, it is that three of eight stated no fee figures at all while ranking for queries about which gateway costs less. The most-cited comparison content in this category is not comparing prices; it is comparing adjectives.

The two errors are different species. The stale BYOK threshold is a maintenance failure: correct when written, never re-checked, and it sits on the page currently ranking first for the query. The crypto claim is a research failure: OpenRouter has charged 5% on crypto payments for as long as we have tracked it, and the article asserts the option does not exist.

We are a competitor to OpenRouter, so read our grading with that in mind. Every verdict links to the primary source, the fee figures are quoted verbatim from vendor pages, and the audit date is on the table. Check any row you doubt.

How to read any gateway comparison

Three questions settle most of it. Is there a date on the claim? Undated fee figures in a repricing market are decoration. Does it name the fee basis? A percentage on credits, a markup on tokens, and a flat monthly fee are three different structures and the difference grows with your spend. Does it state what the author sells? Every page in this category including ours has an interest.

Then verify the two or three numbers your decision actually turns on. It takes ten minutes on the vendors' own pricing pages, and as this audit shows, it is ten minutes more than most published comparisons spent.

Questions, answered