'open weightsA model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.Full definition →' hides differences in license, training data and methodology disclosure. A standardized rubric helps engineers judge what they can host, fine-tuningTaking a trained model and training it a bit more on your own examples so it gets better at one specific job.Full definition → or replicate.
Key takeaways · AI-distilled
Artificial Analysis' December 2025 Openness Index assessed access and licensing for weights, training data and methodology. Available weights alone were insufficient to establish that a model could be fully analyzed or replicated.
The December 2025 Openness Index used a qualitative rubric summed into a score. Artificial Analysis explicitly declined to prescribe an optimal score, recognizing safety, competition and data-disclosure tradeoffs.
At the December 2025 Openness Index launch, OLMo led at 89 and no evaluated model reached the maximum. Artificial Analysis attributed OLMo's result to permissive releases across weights, training data and code, beyond weights alone.
Terms in this piece · Glossary
pretraining — The first, biggest phase of building a model: training it on enormous amounts of text so it learns language, facts, and reasoning in general.
open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
fine-tuning — Taking a trained model and training it a bit more on your own examples so it gets better at one specific job.
