DiscoveryRSI: Open Framework for Recursive Scientific Hypothesis Generation Outperforms Top Models

DiscoveryRSI, released by Princeton's Ling Yang team with 16 partner institutions, uses multi-agent evolutionary search and recursive self-improvement to generate testable scientific hypotheses, achieving a 6.3% hit rate that beats GPT-6 and Claude Opus 5, with an open-sourced 27B model.

2026-10-17 ~ 2026-10-19 · 3 related posts

• Gen-Verse Releases DiscoveryRSI-27B for Recursive Scientific Hypothesis Development — jacek2023 · 2026-10-17
• DiscoveryRSI: recursive multi-agent evolution for scientific hypothesis discovery — PhAILabs · 2026-10-19
• DiscoveryRSI paper: 27B model's hypotheses hit 6.3%, beating GPT-6 and Claude Opus 5 — Afinetheorem · 2026-10-19