NorizonAI raises hundreds of millions RMB for real-time recursive self-improvement AI

智东西 · wechat · 2026-11-19

Beijing-based AI startup NorizonAI announced a nine-figure RMB angel and angel-plus round backed by 5Y Capital, Sequoia China, Legend Star, Zhongguancun Science City Fund and Gobi Ventures.

The company focuses on Real-Time Recursive Self-Improvement (RSI) — letting LLMs keep learning after deployment and update their own parameters. Its core SHINE hypernetwork generates parameter updates for a target model in a single forward pass, aiming to bake new knowledge into weights rather than relying on longer context windows or RAG. It claims better new-knowledge learning and general-task performance than fine-tuning baselines while mitigating catastrophic forgetting.

Founder Zhang Muhan is an assistant professor at Peking University's Institute for AI and previously worked on large-scale graph ML at Meta AI. The 20+ person team includes researchers from PKU, Tsinghua and MIT.

Original post →

More from Venture

Venture channel →