Apex Intelligence is building foundation models designed to act as scientific researchers: take a research problem, form hypotheses, use tools to run computational experiments, test results and revise the approach. Scientists and R&D teams working in AI, chip design, molecular development and quantitative research are the intended users. The Beijing startup says it has completed angel and angel-plus rounds totaling nearly $50 million.
The company calls its approach recursive self-improvement. Through the middle and later stages of model training, Apex aims to teach its systems to generate ideas and carry them through a loop of experimentation, validation and iteration rather than treating the first output as a finished answer. Apex’s public site currently offers a waitlist for Apex Research rather than broad commercial access.
Apex says its system has improved decisions made before AI-model training, developed training strategies and optimized the computing kernels beneath the models. It also claims the system produced conference-level AI research and a complete proof of a decades-old mathematical conjecture. Those are company-reported research results, not evidence of outcomes delivered for paying customers.
The intended path to a useful product runs through research settings where results can be measured. In chip design, molecular R&D and quantitative-strategy development, the model could connect to specialized tools, observe an experiment’s result and use that signal to begin another cycle. Apex’s distinguishing bet is that this loop can improve the research process itself, rather than merely retrieve existing knowledge or draft an answer.
IDG Capital, LinkX Capital and XtalPi co-led the angel round, while Zhongguancun Science City Fund, SCGC and Shanghai Engine Fund co-led the angel-plus financing. Decent Capital, SEE Fund, Monad Ventures and Winsoul Capital also participated. Apex plans to invest in foundation-model training, computing capacity, research-trajectory data and hiring; the announcement did not provide a precise dollar total or valuation.
The capital gives the company, founded in June 2026, resources to move beyond isolated research demonstrations and build the infrastructure needed for repeated experiments. Its next business test is whether externally reviewed results can become workflows that laboratories and engineering teams trust enough to adopt—and whether a waitlisted research system can develop a clear route to revenue.