Fully Open Meditron: An Auditable Pipeline for Clinical LLMs
The Fully Open Meditron project introduces an auditable pipeline for clinical decision support systems using large language models. This initiative aims to enhance transparency and reproducibility in medical AI by providing a complete training stack. The results indicate that fully open models can achieve state-of-the-art performance while maintaining auditability.
- ▪Fully Open Meditron is the first fully open pipeline for building LLM-based clinical decision support systems.
- ▪The project combines a clinician-audited training corpus with a reproducible data construction and training framework.
- ▪Apertus-70B-MeditronFO achieved a new state-of-the-art performance with a 6.6 point improvement over its base model.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.16215 (cs) [Submitted on 15 May 2026] Title:Fully Open Meditron: An Auditable Pipeline for Clinical LLMs Authors:Xavier Theimer-Lienhard, Mushtaha El-Amin, Fay Elhassan, Sahaj Vaidya, Victor Cartier-Negadi, David Sasu, Lars Klein, Mary-Anne Hartley View a PDF of the paper titled Fully Open Meditron: An Auditable Pipeline for Clinical LLMs, by Xavier Theimer-Lienhard and 7 other authors View PDF HTML (experimental) Abstract:Clinical decision support systems (CDSS) require scrutable, auditable pipelines that enable rigorous, reproducible validation. Yet current LLM-based CDSS remain largely opaque.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
