Preprints.ai runs multi-agent AI peer review against new preprints from bioRxiv and medRxiv. Eleven specialist agents review each preprint independently — four integrity agents (methodologist, statistician, ethics, validity), five domain agents, a devil's advocate and a positive-evidence reviewer — then deliberate to produce a 10-tier Publication Fit score anchored to real journals, giving a quality signal before formal review exists.
Benefits & outcomes
Deterministic checks run before any LLM: paper-mill detection, statcheck p-value verification, GRIM tests, data availability, retraction cross-referencing.
~$0.27 and 3–7 minutes per paper; ~85% typical agreement between agents.
Borderline or low-agreement cases arbitrated by Claude Opus with the full agent panel as context.
Notes & quotes
Site is explicit that output is a "machine-generated indicator, not peer review" — grading calibrated to eLife reviews.
Via Scalene: https://scalene-peer-review.beehiiv.com/
Publisher / company: Preprints.ai
Date of mention: 18/07/2026
Source: https://preprints.ai/