Independent AI index
for professionals

Home / Explanation

How we measure

Two kinds of scores, always side by side: independent quality and our own judgement. Where we give advice, per profession or in a comparison, they carry equal weight.

Quality, independently measured

From LMArena: users are shown two anonymous answers and choose the better one. Thousands of such duels produce an Elo score, which forms the index's ranking. A difference of around 20 points is noticeable; a few points is noise.

Editorial score, out of 10

Our grade for what a professional needs: suitability for the work, price including what you have to buy in practice, and privacy.

EloLMArena

Quality from blind duels between models: general and code.

arena.ai · retrieved 9 November 2026
OpenRouteropenrouter.aiPer-token prices and context window sizes, live.
HuggingFace and GitHubhuggingface.coHow popular an open model is with developers.
Hacker Newsycombinator.comWhat the professional community says about it, praise and criticism alike.
1
Source recordedEvery fact on the site has a verifiable source.
2
Provider checkedProduct page and terms of the provider itself.
3
Price and GDPR recordedPer plan, and what the provider says about your data.
4
Checked every morningIf something changes, the new date is shown alongside it.

What we don't do

  • Sell paid placements
  • Let providers influence rankings
  • Hide affiliate links: if they ever appear, they are clearly marked

Methodology: the GDPR profile

The GDPR profile summarises what a provider itself publishes about data protection. We verify against public documentation and date every check.

A tool for your shortlist. Not legal advice and not an audit. "Unknown" literally means: the provider does not disclose it.

The full methodology: GDPR profile, Elo, editorial score and checks

StrongSeveral safeguards published: EU hosting or data residency and no training on your data, and/or self-hosting possible. For tools: data in the EU, certification and a Dutch/EU establishment.
GoodTwo of the safeguards above published.
LimitedOne safeguard published; the rest unknown.
UnknownThe provider publishes (almost) nothing about data protection.
CautionThere is a concrete, documented point of concern, for example an investigation by a regulator or a licence restriction.

Terms in plain language

No technical background needed. The terms used on this site, with how much they matter in practice.

Elo
A ranking score from millions of blind 1-versus-1 comparisons by real users (LMArena). Higher = more often chosen as the best answer. A difference of ~20+ points is noticeable in practice; a few points is noise.
context window
How much text a model can hold 'in its head' at once, measured in tokens. Large = you can supply whole case files or long documents in one go.
tokens
Chunks of text (roughly ¾ of a word) that AI counts in. Prices and limits are measured per token, 1 million tokens is roughly 750,000 words.
API
A way to build the model into your own software or website, rather than using the chat app. You then pay per token instead of a fixed subscription.
on-premise
The model runs on your own servers rather than at the provider. Important when sensitive data must not leave your organisation (GDPR).
modalities
Which kinds of input and output the model handles: text, image, audio or video.
editorial score
Our own rating (out of 10), a weighting of suitability, price and privacy, alongside the independent Elo.
tokens/sec
How fast a model types out an answer. Higher = faster; noticeable with long answers.
data processing agreement (DPA)
A contract in which the AI provider sets out what happens to your data. Required under the GDPR as soon as you process personal data; business subscriptions almost always include one.
open weights
The model itself can be downloaded and run on your own hardware (such as Llama and DeepSeek). Maximum control over your data, but you arrange the servers and security yourself.
hallucination
An AI answer that sounds convincing but is factually wrong. Every AI sometimes does this, which is why a professional should always be involved in decisions that matter.
prompt
The instruction or question you give an AI. The more concrete the prompt (context, example, desired format), the better the answer.
GDPR profile
Our summary of what a provider publishes about data protection, from Strong to Caution. See the methodology at the bottom of this page.