Stay connected with us on X, Discord, and LinkedIn to stay up to date with future analysis

Openness Index Methodology

Overview

The Artificial Analysis Openness Index is a composite metric that measures the degree to which AI models are openly available and transparently documented. It covers both the availability of model weights and the transparency of the underlying data and methodology used to create the model.

Each model is assessed based on the full set of public first-party information available. Where models are derived from a third-party base model, they may be constrained by the licensing or limited disclosure of the upstream model. For incremental or update releases, only disclosures explicitly about the new release are considered (including allowing model creators to declare which components remain consistent with an earlier release).

A detailed methodology specification is available to download as a PDF.

Index Composition

1. Model availability
Weights
Access
0Closed weights, no API
1Closed weights, API limits token visibility
2Closed weights, API available
3Open weights
License
0Closed weights or no commercial use
1Commercial use, attribution required
2Commercial use, no attribution required
3Commercial use, no attribution required, no meaningful limitations
2. Model transparency
Data:Pre & Post Training(score represents average across each)
Access
0No or limited disclosure
1Partial data source detail and categorization disclosed
2Full data mix disclosure, substantial data shared¹
3Full data shared
License (most restrictive)
0No commercial use/no substantial data shared
1Commercial use, attribution required
2Commercial use, no attribution required
3Commercial use, no attribution required, no meaningful limitations
Methodology
Disclosure
0No or limited disclosure
1Model architecture disclosure
2Limited general technical disclosure
3Full technical details disclosed
License (most restrictive)
0No code disclosed/released
1Frameworks disclosed, openly available for commercial use
2End-to-end training pipeline code or guide released
3End-to-end training pipeline code or guide released, and commercial use allowed

Each component is scored on a 0–3 qualitative scale based on the best-fitting openness archetype, with each model assessed based on the full set of public first-party information available.

Where models are derived from a third-party base model, they may be constrained by the licensing or limited disclosure of the upstream model. For incremental/update releases, we only consider disclosures explicitly about the new release (including allowing model creators to declare which components remain consistent with an earlier release).

Score Calculation

The final Openness Index score is derived as follows:

  • Data components are scored separately for pre-training and post-training, then averaged to give a combined data score (up to 6 possible points across Access and License).
  • All component scores are summed, for a maximum raw score of 18.
  • The raw score is normalized to a 0–100 scale: