Clinical care, health‑system operations and biomedical discovery all rely on extracting reliable answers from patient records, yet applying diverse clinical questions across whole populations is computationally expensive. We introduce Clinical System One (CS1), a 4‑billion‑parameter decision model that delivers fast, typed judgments from clinical records and supports two agentic setups where CS1 collaborates with a generalist language model. CS1‑4B was trained on 12.4 million question‑answer pairs from 3.6 million patients across a multi‑hospital system, combining structured data keys, observed outcomes and language‑model supervision over clinical notes, diagnostic reports and patient timelines. The model produces multiple typed answers in a single forward pass without task‑specific fine‑tuning, and each answer carries a probability that can route uncertain cases to human review. On 963,408 held‑out patients, CS1‑4B outperformed open decision models up to 27 billion parameters on 13 of 14 tasks. Across 21 external data sets its aggregate performance approached the strongest 27‑billion‑parameter model while delivering 6.4‑fold faster inference under matched workload. Integrating CS1‑4B with a language‑model agent yielded cited answers over complete multi‑year patient histories and screened a patient population against the eligibility criteria of 17 registered clinical trials. The approach placed four times as many judged‑eligible patients among the top‑20 candidates compared with keyword search, and used 690‑fold fewer language‑model tokens than a model that reads every record. These findings support a division of labour in clinical AI: small decision models handle repeated record reading, while generative models formulate questions and synthesize evidence.
Review