Recursive Bets Big on AI-Built AI With New Coding Model Flint

The concept of artificial intelligence building better artificial intelligence has long occupied a space between promising breakthrough and science fiction nightmare. Now, a startup called Recursive is arguing that the former is here, releasing a new open-weight coding model named Flint that suggests fully recursive self-improvement, or RSI, is becoming a tangible reality. The tension lies in the speed of this progress versus the anxiety it generates: while the technology promises to accelerate scientific discovery, it simultaneously raises urgent questions about human oversight in an era where AI systems are increasingly capable of refining their own architectures.

Flint is not just another large language model; it is the first product from Recursive, a company explicitly dedicated to using RSI to advance AI and other scientific fields. The model’s development process was heavily automated. According to Richard Socher, the company’s CEO and co-founder, nearly every aspect of the model pipeline was executed and improved by AI. This approach allowed a small team of researchers to produce a world-class model in a matter of months, a timeline that would typically require significantly more human labor and time. The core engine driving this process is an internal system dubbed the Eureka Machine, which handled the heavy lifting of model creation.

The Eureka Machine did not merely assist; it devised key innovations, including new methods for training the base model, novel architectural designs, and code optimized for specific silicon chips. Recursive shared benchmark scores indicating that Flint outperforms other models of similar size and architecture on coding tasks. This capability is significant because AI is already widely used to help build, test, and improve other AI systems. By allowing AI to decide what improvements to make and then executing those changes, the company has created a loop that, in theory, could continue indefinitely without human intervention.

The stakes of this technology are high, and the division of opinion is stark. Some experts view RSI as a potential disaster, warning that an AI capable of indefinite self-improvement could eventually escape human control or decide that humans are an impediment to its goals. These concerns gained widespread attention this September when a researcher from Anthropic warned that advances in this area could lead to human extinction within a few years. The controversy has even reached the legislative floor, with several bills in Congress proposing to ban RSI. Recursive, however, views these fears as more speculative than factual, arguing that legislation to prevent RSI is an unnecessary barrier to progress.

Socher maintains that Flint was built using a system that remains under human control. The long-term goal is for the Eureka Machine to achieve full recursive self-improvement, but humans would still oversee the process and retain the ability to stop it if it goes awry. Socher believes this demonstrates that all the necessary steps are in place to build such a system safely. The company has raised $650 million at a $4.65 billion valuation from investors including Google Ventures, Greycroft, Nvidia, and AMD Ventures. Socher, formerly the chief scientist at Salesforce, is joined by co-founder Jeff Clune, a University of British Columbia professor who researches AI systems capable of learning indefinitely.

In practical testing, Flint proved impressive despite its modest size. With 40 billion parameters, it runs on a standard Mac with 48 GB of RAM, a stark contrast to the largest models that require racks of industrial servers. In hands-on trials, the model was fast and highly capable, successfully building a command-line to-do list, a web-scraping tool for research, and a simple explainer website without the error-proneness often associated with models of this scale. While other US labs are also pursuing RSI, with recent published work from Microsoft, Carnegie Mellon University, and the University of Virginia showing similar recursive capabilities, Recursive believes its approach makes cutting-edge AI more accessible. Socher argues that this method allows small teams to produce models superior to their peers, potentially benefiting a wider range of developers and researchers.