NORIZON: Closes RMB 100M-Level Angel and Angel-Plus Rounds

On 19 November 2026, Norizon AI closed RMB 100 million-level angel and angel-plus rounds from 5Y Capital, Sequoia China and others, to advance its RSI model.

NewTimeSpace News: On 19 November 2026, Norizon AI (Beijing Norizon AI Technology Co., Ltd., unlisted) announced angel and angel-plus rounds totalling several hundred million yuan, with investors including 5Y Capital, Sequoia China, Legend Star, ZGC Science City Fund and Gobi Partners. No lead investor was designated; the institutions invested in parallel. FlowCapital acted as financial adviser rather than an investor. The exact amount and post-money valuation were not disclosed. Proceeds will mainly fund research and product rollout for its self-developed RSI model. Registered in Haidian district, Beijing in June 2026, the company's name is drawn from a phrase meaning that life is bounded while knowledge is not.

Norizon focuses on Real-Time Recursive Self-Improvement, making parameter-based continual learning its core route, so that a model can keep absorbing new knowledge and internalising experience after deployment. The company argues that larger context windows, retrieval-augmented generation and external memory do not write new knowledge into model parameters. Its approach uses a pre-trained hypernetwork to generate parameter updates for the base model directly, completing light adaptation in a single forward pass and bringing the cost of one update close to that of inference. A hypernetwork is a network that generates part or all of another network's parameters.

The company says early validation on new-material internalisation showed that an adapted model could answer related questions without reading the source text, and that its hypernetwork outperformed fine-tuning both in learning new knowledge and in retaining general-task performance. These are company-reported experimental results and have not been independently verified. Further work covers enlarging the hypernetwork, lengthening each internalisation, stability across millions of tokens of continuous updates, and mechanisms for active judgment and safe rollback.

Founder and chief scientist Dr Zhang Muhan is an assistant professor and doctoral adviser at Peking University's Institute for Artificial Intelligence. He studied at Shanghai Jiao Tong University's IEEE pilot class, earned a doctorate at Washington University in St Louis and was a research scientist at Facebook AI, now Meta AI. The team has more than 20 members from Peking University, Tsinghua, the University of Hong Kong, Fudan and MIT, including olympiad and ICPC gold medallists and researchers in post-training and agents. Norizon is exploring AI-agent experience learning, enterprise model adaptation and personalised models, with both local deployment and cloud services.

×
Share to WeChat

Open WeChat, use the "Scan", and share to my Moments.