What we will change
- Rate limits, quotas, or credit amounts that have moved
- A model list that has rotated on since our snapshot
- Retention, training, or routing terms we read too conservatively
- A tier or privacy classification that no longer matches the offer
- An offer that launched, expanded, narrowed, or ended
- A wrong name, endpoint, operator, or link
What we need from you
One first-party link that settles it: current pricing, documentation, rate limits, terms, or a live models endpoint. Say which record and which field is wrong, and what it should say instead.
A launch post or a marketing page is a lead, not proof. Where an announcement and the governing terms disagree, the terms win — that rule is what keeps the catalog defensible, and it applies to corrections in your favor exactly as it applies to ones against you. If the decisive number lives only behind a login, say so and we will label it dashboard-only rather than guess at it.
How a correction lands
We re-audit the record against first-party sources, update it in the source dataset, and re-date its snapshot. The site is rebuilt from that file, so the change reaches every page at once — the provider page, the catalog ordering, and the counts on the homepage. The prior claim stays in the record’s history, so the correction is auditable rather than silent.
What we will not do
- Remove a record because it is unflattering
- Call an offer private without plan-specific evidence
- Take a marketing claim over the governing terms
- Drop a watchlist or shame entry on request alone
The shame page is not a punishment. It records that no free hosted inference was verifiable on the date we checked. Publish a genuinely free API — or point us at the one we missed — and the record moves into the catalog on the next audit.
Where to reach us
The Discord is the fastest route and the one we watch. Post the record, the field, and the link; corrections and provider questions are both welcome there, and so is an argument that we got a classification wrong.