Skip to content

Start here: what you are deciding

The AI Ownership Index rates how much you own an open model or an inference provider. Each entry ends in one ownership level, from full to none, and shows the four factors behind it. The level is floor-weighted: the weakest factor caps it.

Whether anyone can stop you using or changing it. An open-weight model you download cannot be withdrawn, so the licence decides. Some licences are permissive (MIT, Apache-2.0). Others add conditions such as a monthly-active-user threshold, a revenue cap or a limit on certain uses. Each entry states its licence and links the text. For a provider, this factor also covers how portable the service is and what it costs.

What you can see: the weights, the training data and code, the evaluations and the terms. A model that publishes only its weights rates lower than one that also publishes its data and recipe. For a provider, the question is whether the binding terms are published and legible.

Whether it runs dependably and whether misuse is controlled. This factor combines operational readiness and safety. How capable a model is affects the AOI score and leaves this factor unchanged.

Whether using it hands your data to someone else. A self-hosted model runs on your hardware and sends nothing back. With a hosted provider, the Terms and Privacy Policy decide whether it keeps your prompts, trains on them and owns the outputs, and where processing happens. The index reads those documents and records where they disagree with the marketing.

Beside the level, each entry shows an AOI score from 0 to 100 and a grade from A to F, built from seven weighted dimensions. Trust dimensions outweigh capability. The score is an input to the verdict, and the level is the verdict.

A self-hosted model costs hardware, and no vendor can raise a per-token price. Hosted providers charge per use, and each provider entry records its pricing.