NEWS · MODELS · #2320
Clinical System One (CS1-4B): 4B-parameter decision model for clinical records (arXiv:2610.27099v1)
Researchers introduce Clinical System One (CS1-4B), a 4-billion-parameter decision model trained on 12.4 million Q&A pairs from 3.6 million patients across a multi-hospital system that produces typed judgments and probabilities from full clinical records in a single forward pass. CS1-4B outperformed larger open decision models on most held-out tasks, achieved comparable aggregate performance to a tested 27B model with 6.4× faster inference, and when paired with a language-model agent enabled cited answers over multi-year histories and much more efficient trial-eligibility screening than keywords or full-record LLM reads.
KEY POINTS
- Researchers introduce Clinical System One (CS1-4B), a 4-billion-parameter decision model trained on 12.4 million Q&A pairs from 3.6 million patients across a multi-hospital system that produces typed judgments and probabilities from full clinical records in a single forward pass.
- CS1-4B outperformed larger open decision models on most held-out tasks, achieved comparable aggregate performance to a tested 27B model with 6.4× faster inference, and when paired with a language-model agent enabled cited answers over multi-year histories and much more efficient trial-eligibility screening than keywords or full-record LLM reads.
- Shows that a relatively small, task-specialized decision model can read patient records repeatedly and cheaply while matching or exceeding larger models on many tasks, enabling scalable clinical workflows and cheaper LLM integration.
WHY IT MATTERS
Shows that a relatively small, task-specialized decision model can read patient records repeatedly and cheaply while matching or exceeding larger models on many tasks, enabling scalable clinical workflows and cheaper LLM integration.