---
title: "Free Inferencing methodology"
description: "How Free Inferencing discovers, verifies, and classifies free hosted model APIs."
canonical_url: "https://freeinferencing.com/methodology/"
md_url: "https://freeinferencing.com/methodology.md"
last_updated: "2026-09-19"
---

# Free is a claim. We make it prove itself.

> The catalog is generated from a versioned YAML audit, not affiliate copy or a scraped listicle.

## What qualifies

A provider must run the model remotely and expose a callable API. The catalog covers text generation, embeddings, reranking, speech, image, safety, and user-deployed model endpoints when hosted compute is genuinely covered by the offer.

## What does not qualify

- A free web chat with no developer API
- Free weights that require self-hosting
- A free gateway that still bills the upstream model
- Mocked sandbox responses
- Stale launch posts contradicted by current billing documentation

## Evidence order

Live official catalogs come first, followed by current pricing, billing and limits documentation, product pages, changelogs, official repositories, company announcements, and reputable corroboration. Search snippets and third-party trackers are leads, not proof.

## Verification boundary

Most endpoints require an account. A public audit cannot prove an authenticated generation or a zero-dollar bill unless the provider offers a credential-free call. Dashboard-only gaps and conflicting first-party claims stay visible on the watchlist.

## AI-assisted maintenance

AI agents research providers, classify offers, maintain the structured source dataset, and generate the site's code and editorial surfaces. That workflow does not make an unsupported claim acceptable: every material finding keeps its primary evidence attached, uncertainty remains explicit, and readers can inspect the same sources and submit corrections.

## Prompt and response data

Privacy is a separate audit of service-specific terms, privacy policies, acceptable-use rules, and developer or data-processing agreements. "No model training" does not necessarily mean no logging, human review, routing, or broader product improvement. Missing documentation is never treated as a privacy promise.

## Re-verification

A verified record is a dated finding, not a permanent badge. Model catalogs, quotas, billing requirements, eligibility, and governing terms are re-opened when a source changes, a correction arrives, or the snapshot is refreshed. Records move between current, watchlist, and excluded groups when stronger evidence changes the bounded claim; a dashboard-only value remains uncertain until it can be reproduced or documented publicly.

## Corrections

Every material claim keeps its first-party sources. Send a current provider link through the [corrections page](/contact.md) when a limit, model, policy, or offer changes.

## Sitemap

See the full [semantic sitemap](/sitemap.md) for every page and markdown mirror.