The Pulse
Apex Intelligence Raises Nearly $50M for Self-Evolving Models
The Beijing startup says it will use the financing to develop foundation models designed to improve their own research processes, while its claims about benchmarks and mathematical work remain unverified outside company materials.
AI.info Team ·
“I have always believed that models taught by humans are ultimately limited by humans themselves. When we align them to human preferences, we lock their intelligence ceiling at our own level. No matter how much compute and data we add, we are only making them better at retaining existing knowledge. The limits of true intelligence can only be defined by the objective world. The first generation of large models replaces white-collar workers; embodied AI replaces blue-collar workers; and ASI will replace humanity's most accomplished scientists. That moment will come sooner or later. Our job is to bring it closer.”
Yongchao Chen, founder of Apex Intelligence
Beijing-based Apex Intelligence says it has raised nearly US$50 million across angel and angel-plus funding rounds to develop self-evolving foundation models. The company announced the financing on September 16, 2026, after being founded in June.
IDG Capital, LinkX Capital and XtalPi co-led the angel round, with Decent Capital, SEE Fund, Monad Ventures and Winsoul Capital also participating. Zhongguancun Science City Fund, SCGC and Shanghai Engine Fund co-led the angel-plus round. Apex Intelligence did not disclose the split between the two rounds, its valuation or the size of individual checks.
From AI Assistants to AI Researchers
Apex Intelligence describes its technical focus as recursive self-improvement, or RSI: models improving themselves through repeated cycles of generating ideas, testing them, checking the results and using the evidence to shape subsequent work.
The company says its goal is to move AI beyond assistive tools and toward systems that can act as co-researchers. It is focusing on scientific research and AI-for-AI work, with training methods that use mid-training and post-training to encourage models to generate ideas and carry out hypothesis formation, experimentation, validation and iteration.
Potential applications include chip design and simulation, molecular research and quantitative strategy development, according to the announcement. The company says these areas are suited to its approach because their research processes can be structured and their outcomes measured.
Company Claims About Early Results
Apex Intelligence says its system has autonomously produced research it considers suitable for acceptance at leading AI conferences, with capabilities it compares with those of an AI PhD student. It also says the system has made more accurate decisions before training, developed improved strategies during training and optimized underlying GPU kernels.
The release refers to results across SimpleTES, NanoChat Autoresearch, GPUMode TriMul and MLS-Bench. Apex says its system set new records or approached the best publicly reported results on those tasks, outperforming or matching work associated with Recursive Superintelligence, Tencent Hunyuan, Stanford and NVIDIA.
Those statements are claims from Apex Intelligence. The release does not provide an independent audit, peer-reviewed evidence or outside verification of the reported performance.
In mathematics, the company says its system produced a complete proof of the majorization conjecture, which it describes as having remained open for more than three decades. The announcement does not identify a peer-reviewed publication or independent mathematical verification for that proof, so the claim should be treated as an assertion by the company rather than an established result.
Funding and Recruitment
Apex Intelligence says it will use the financing for foundation-model development, computing infrastructure, recursive model training, research-trajectory data infrastructure and hiring. Its long-term target is to build self-evolving foundation models whose combined research capabilities match and eventually surpass those of 100,000 top-tier AI scientists across disciplines.
The company characterizes that target as a long-term vision and says it is working toward it incrementally. The funding announcement describes the objective, not a current capability.
Apex Intelligence plans to visit Harvard University, MIT, Yale University and Boston University from September 17 through September 23 to meet students, researchers and academic research groups for recruitment and research-collaboration discussions. The company lists Beijing as its base and says it is hiring across research and engineering roles.
Chen founded Apex Intelligence and is identified in the announcement as an assistant professor at Tsinghua University's School of Artificial Intelligence. The company says its core team includes researchers from Tsinghua University, Peking University, Harvard and MIT.
The startup's immediate challenge is to show that its reported results can be reproduced and extended beyond tasks selected by its own researchers. For now, the financing supports an ambitious effort to develop models that participate in research rather than only execute objectives defined by people.