actAVA AI's Cura 1T tops healthcare benchmarks with self-evolution
A 1-trillion parameter model that improves itself through targeted data refinement across five key healthcare tasks.
actAVA AI has released Cura 1T, a 1-trillion parameter LLM purpose-built for agentic healthcare applications. Unlike general-purpose models, Cura 1T is trained through a novel human-gated self-evolution loop. In each evolution round, a training agent plans a target capability, trains the model, evaluates benchmark trajectories, and refines the data mixture based on observed failures. This data-centered approach improves the model with targeted synthetic and curated examples rather than a single generic medical-data update, preventing performance degradation across different healthcare tasks.
Cura 1T covers the full spectrum of healthcare use cases: patient consultation, clinical reasoning over text and images, interactive diagnosis, and electronic health record (EHR) tool use. Across the healthcare evaluation suite, it ranks at or near the top among frontier baselines, including GPT-4 and Claude. Remarkably, it remains competitive on out-of-domain reasoning and general agentic benchmarks, suggesting the specialized training does not sacrifice broad capabilities. The model is open-source, with code and documentation available on GitHub and Hugging Face.
- Human-gated self-evolution loop enables continuous improvement across multiple healthcare tasks without degrading others.
- Handles patient consultation, clinical reasoning (text+images), interactive diagnosis, and EHR tool use.
- Ranks at or near top among frontier baselines on healthcare suite, competitive on out-of-domain reasoning benchmarks.
Why It Matters
Specialized healthcare AI that safely reasons, diagnoses, and executes workflows could revolutionize clinical decision support and patient care.