Researchers introduce Cura 1T, a healthcare-specialized LLM trained via a human-gated self-evolution loop

Researchers at actAVA AI describe Cura 1T, a healthcare-specialized large language model designed to unify high-stakes clinical capabilities such as patient consultation, clinical reasoning over text and images, interactive diagnosis, and electronic health record (EHR) tool use. The team argues that existing specialized LLMs often fail to cover these use cases together, and that narrow updates for one task can degrade performance on others.

To address this, Cura 1T is trained through a human-gated self-evolution loop. In each round, a training agent selects a target capability, trains the model, evaluates benchmark trajectories, and then refines the data mixture based on observed failures. This process emphasizes targeted synthetic and curated data rather than a single generic medical-data update, aiming to improve targeted weaknesses without introducing regressions.

The authors report that across a healthcare evaluation suite, Cura 1T ranks at or near the top among frontier baselines. They also state that the model remains competitive on out-of-domain reasoning and agentic benchmarks, suggesting potential generalization beyond clinical settings.

The preprint includes links to a model page, documentation, and a GitHub repository, indicating that the artifacts are intended to be publicly accessible for further evaluation and use.