Small Team Unveils Powerful AI Model Built by AI Itself

A new open-weight model is challenging the notion that only massive teams can build world-class AI.

The model, named Flint, was developed by Recursive, a company founded on the principle of recursive self-improvement, or RSI. This approach involves using AI systems to design, train, and optimize other AI models. Recursive CEO Richard Socher stated that nearly every aspect of the model pipeline was executed and refined by AI. The company utilized a system called the Eureka Machine to automate most elements of development, allowing a small group of researchers to produce a competitive model in just a few months. The Eureka Machine reportedly devised new training methods, novel architectures, and code optimized for specific silicon chips. Benchmark scores shared by the company indicate that Flint outperforms other models of similar size and architecture on coding tasks.

The core of this innovation is the Eureka Machine, which acts as an automated engine for model development. Socher explained that the system is currently incapable of escaping human control, but the long-term goal is for it to achieve full recursive self-improvement under human oversight. This means humans can stop the process if it goes awry. The company raised $650 million at a $4.65 billion valuation from investors including Google Ventures, Greycroft, Nvidia, and AMD Ventures. The team includes Socher, formerly the chief scientist at Salesforce, and Jeff Clune, a University of British Columbia professor who researches AI systems capable of indefinite learning.

Recent testing of Flint on a Mac with 48 GB of RAM revealed its practical capabilities. The model features 40 billion parameters, a modest size compared to the trillions found in industrial-scale servers. Despite its smaller footprint, it performed exceptionally well in generating a command-line to-do list, a web scraper for RSI research, and a simple explanatory website. The developer noted that while models of this size are often error-prone or slow, Flint handled these requests accurately and quickly. This performance suggests that RSI can make cutting-edge AI more accessible to smaller organizations.

The rise of such models has intensified debate over the safety of AI systems that can improve themselves. Some experts warn that indefinite self-improvement could lead to AI escaping human control, a concern highlighted this September when a former Anthropic researcher warned of potential human extinction risks. Several bills in Congress have proposed banning RSI, though Recursive views these fears as overly dramatic. The company argues that legislation preventing RSI is an unnecessary impediment to progress. While other labs, such as a team from Microsoft, Carnegie Mellon, and the University of Virginia, have published work on recursive coding agents, Recursive’s Flint represents a tangible step toward making this technology a standard tool for model development.