An AI agent for treatment reasoning over a biomedical tool universe

Short summary

ATHENA-R1 is an AI agent for treatment reasoning trained via reinforcement learning over 212 biomedical tools, achieving 94.7% accuracy on drug reasoning—17.8 points above GPT-5. Validated by rare disease experts and tested on 5.4M patient records with clinically meaningful adverse-event predictions.

• •AI agent trained with RL to reason over FDA-approved drugs and biomedical tools
• •94.7% accuracy on drug reasoning, 82.9% on patient treatment cases—outperforms GPT-5 by 17-18 points
• •Expert validation from rare disease organizations and real-world EHR data testing

Generated with AI, which can make mistakes.