Apex Intelligence Claims $50M for Self-Evolving Foundation Models

The technical claim matters more than the check size. Apex Intelligence says it is building foundation models that can recursively improve themselves, using mid-training and post-training to guide models to generate research ideas and autonomously carry out self-directed cycles. For the AI research community, that is a red-hot but unproven frontier—if it works, it would shift models from executing known tasks to generating novel scientific directions.

Apex Intelligence, a Beijing-based startup founded in June 2026, announced on September 16 that it has completed its angel and angel-plus funding rounds, raising nearly US$50 million in total. The company is building what it calls self-evolving foundation models, meaning models that can improve themselves through Recursive Self-Improvement, or RSI, with a primary focus on scientific research and discovery. The information comes from a press release distributed through PR Newswire and syndicated by AOL; no independent reporting has yet confirmed the figures or the technical claims. As such, the round, investor commitments, and product milestones should be understood as company assertions rather than independently audited facts.

“

The angel round was co-led by IDG Capital, LinkX Capital, and XtalPi, with participation from Decent Capital, SEE Fund, Monad Ventures, and Winsoul Capital.

The financing structure is unusual for a company that is only about three months old. The angel round was co-led by IDG Capital, LinkX Capital, and XtalPi, with participation from Decent Capital, SEE Fund, Monad Ventures, and Winsoul Capital. The angel-plus round was co-led by Zhongguancun Science City Fund, SCGC, and Shanghai Engine Fund. The inclusion of XtalPi, a listed AI-driven drug discovery company, and Zhongguancun Science City Fund, tied to Beijing's high-tech district, suggests strategic alignment with both life-science applications and Chinese state-linked innovation policy. That syndicate composition may matter as much as the dollar amount: it combines venture capital, corporate strategic capital, and quasi-public funding into an early-stage AI research bet.

Apex Intelligence's technical thesis is equally ambitious. The company argues that most AI today remains focused on engineering and execution, while true scientific breakthroughs still depend heavily on humans. Apex wants to build a 'native research engine' that can move beyond accelerating familiar work to actively discovering unexplored directions. It specifically points to mid-training and post-training as the levers for building model capabilities in research and development. In the broader AI landscape, recursive self-improvement is both a research frontier and a safety concern: many labs debate whether models can reliably improve their own capabilities without compounding errors, misalignment, or unintended behavior.