Start here: what you are actually deciding
You do not need to read benchmarks to choose an open model. The decision comes down to how much you really own what you are adopting. This index scores exactly that as the AI Ownership Index (AOI), and it breaks into three plain questions.
Every model and provider page opens with a Bottom line box that answers these three at a glance, and closes with the full, sourced Ownership verdict.
1. Can it be restricted?
Section titled “1. Can it be restricted?”The worry: can someone stop you doing what you want with it, now or later? Closed services can change their terms, add usage caps, or cut you off.
What to look for: an open-weight model you can download removes most of this risk, because the weights are already in your hands. The catch is the licence. Some are fully permissive (MIT, Apache) with no user caps or field-of-use limits. Others attach conditions, for example a monthly-active-user threshold or a restriction on certain uses. Each entry states the licence in one line and links the full text.
Where to check it: the Can it be restricted? line in the Bottom line box, and the Use and modify freely factor in the Ownership verdict.
2. What happens to your data?
Section titled “2. What happens to your data?”The worry: does using this expose or give away your data? This is where open and hosted differ the most.
What to look for: a model you self-host runs on your own hardware, so your data never leaves and nothing is sent back. A hosted provider is different: you are trusting their terms, so the questions become whether they train on your inputs, how long they keep them, whether your data and outputs stay yours, and where it is processed. The index reads the provider’s actual Terms and Privacy Policy, not their marketing, and says plainly when the two disagree.
Where to check it: the What happens to your data? line in the Bottom line box, and the Does not extract your data factor in the Ownership verdict.
3. Will you be locked in?
Section titled “3. Will you be locked in?”The worry: will you get trapped with one provider and be unable to leave?
What to look for: open weights are portable. The same checkpoint runs on many providers or on your own machine, so you can move. With a hosted API, an OpenAI-compatible interface makes moving cheap; a proprietary interface means rewriting your integration, though the underlying open model still runs elsewhere.
Where to check it: the Will you be locked in? line in the Bottom line box, and the Transparency factor in the Ownership verdict.
A note on cost
Section titled “A note on cost”For a self-hosted open model, the cost is your own hardware, not a metered bill that can creep up, and there is no vendor who can raise a per-token price on you. Hosted providers do charge per use; each provider page records the pricing model.
Where this goes next
Section titled “Where this goes next”A later version will show how the best open options compare to the closed frontier, framed on ownership rather than raw performance: how much of a frontier-grade capability you can now have on terms you actually own.