Definition
Plain language
A small AI agent for medicine that answers treatment questions by looking things up in real drug databases instead of trusting its own memory.
As stated in the literature
An 8B-parameter tool-using medical reasoning agent trained via supervised fine-tuning then reinforcement learning inside a 212-tool environment (openFDA, DrugBank, Open Targets), rewarded for evidence-seeking behavior rather than only final-answer correctness; distinct from the scientific-computing framework ATHENA.
Why it matters: It matters because grounding medical answers in checkable sources reduces the risk of a confidently wrong recommendation in a high-stakes setting.
For example, when asked whether a drug is safe alongside a patient's other medications, it queries live drug databases and reports what the evidence says rather than guessing from memory.
Heard on the show
“And the same paper builds a different agent, called ATHENA-R1, that uses a tool on every single case — and beats GPT-5 by about eighteen points on drug reasoning.”Episode 187 — An 8-Billion Agent That Beats Models 80 Times Its Size By Looking Things Up