Reproduce “Cura 1T: Healthcare Foundation Model via Recursive Self-Improvement” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Haolin Chen, Leon Qi, Steve Brown, Deon Metelski, Tao Xia, Joonyul Lee, Qixuan Wang, Kevin Riley and 2 more
Cura 1T improves healthcare model behavior by iteratively refining its training-data mixture in response to benchmark failures. It adds targeted examples while retaining examples of behaviors already handled, with human review. On MedAgentBench, the released model reaches 0.940 task success, above its Kimi-K2.6 base and Claude Opus 4.8, but this benchmark result does not establish clinical safety or readiness for unsupervised care.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.