The abstract is read at the publisher; the summary is JClub's.
Claude Opus 5.5 delivered the highest primary performance across all eight biomedical endpoints, while Jev showed competitive results at under 1% of Opus cost on most datasets.
- Why it matters: Biomedical workflows need strong accuracy but also practical inference cost and latency, and it remains unclear how confidence-based routing can balance performance with expense across varied task types.
- What they did: The study benchmarked three System One models and three general-purpose LLMs on seven biomedical datasets spanning question answering, error detection, structured extraction, and clinical-note assessment, measuring performance, cost, latency, and confidence calibration for routing.
- The result: Three confidence-calibrated SOM-to-LLM routes cut costs by 30.0%–71.7% versus the receiving LLM with modest performance drops, and SOM-to-SOM routing reached 91.1% accuracy versus 82.3% and 82.2% for individual models.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.