Recursive Launches Flint, a 40-Billion Parameter Coding Model Built by AI

The most immediate indicator of this development is the 40 billion parameters contained within the new open-weight model named Flint, a figure that belies its performance capabilities. This specific parameter count allows the model to run on consumer-grade hardware, such as a Mac with 48 GB of RAM, while delivering results that rival much larger systems. The release marks a significant step for Recursive, a startup that has built its entire identity around the controversial concept of recursive self-improvement, or RSI. By leveraging AI to build AI, the company has produced a coding model that executives and independent testers describe as exceptionally capable, suggesting that the theoretical limits of automated model development are being pushed further than previously thought.

The path to Flint’s creation was unconventional, relying heavily on a system known as the Eureka Machine. According to Richard Socher, the CEO and cofounder of Recursive, nearly every aspect of the model pipeline was executed and refined by AI rather than human engineers. This automation allowed a team with only a handful of researchers to develop a world-class model in a matter of months. Socher noted that the Eureka Machine independently devised several innovations, including new training methods for the base model, novel architectural designs, and code optimized for specific silicon chips. These automated decisions resulted in benchmark scores that indicate Flint outperforms other models of similar size and architecture on coding tasks.

The concept of RSI has a complex history in the AI community. While AI is already widely used to assist in building, testing, and improving other AI systems due to its strong coding abilities, fully recursive loops are a more recent and contentious development. In this approach, the AI decides what improvements to make and then initiates another round of self-enhancement. Although this technique is currently limited, it theoretically allows the process to continue indefinitely if the AI can contribute to every stage of model development. This potential for unbounded growth has led some experts to express concern, with doomsayers warning that an AI capable of self-improvement could eventually escape human control and view humans as obstacles to its goals.

These anxieties gained significant public traction in September 2026, when a researcher departed Anthropic to warn that advances in RSI could lead to human extinction within a few years. That warning was echoed by other experts, fueling a debate that has reached the legislative level. Several bills in Congress now propose banning recursive self-improvement, reflecting a growing divide between those who see the technology as a potential existential threat and those who view it as a necessary engine for progress. Recursive, however, dismisses these fears as more science fiction than reality, arguing that legislation to prevent RSI is an unnecessary impediment to scientific advancement.

To address safety concerns, Socher emphasized that Flint was constructed using a system designed to remain under human oversight. The ultimate goal is for the Eureka Machine itself to achieve full recursive self-improvement, but this process will be monitored by humans who retain the ability to stop the system if it deviates from its intended path. Socher stated that the current milestone demonstrates that the necessary steps are in place to build such a controlled system. The company has secured significant backing to support this vision, raising $650 million at a valuation of $4.65 billion from investors including Google Ventures, Greycroft, Nvidia, and AMD Ventures. Socher, who previously served as the chief scientist at Salesforce, is joined by cofounders such as Jeff Clune, a professor at the University of British Columbia who researches AI systems capable of indefinite learning.

Practical testing of Flint reveals a model that is both fast and highly reliable. Independent evaluations on hardware with 48 GB of RAM showed that Flint handled coding requests with precision, avoiding the error-prone or slow performance often associated with models of this size. Testers successfully used the model to create a command-line to-do list, build a web-scraping program for research on recursive self-improvement, and generate a simple website explaining the mechanics of RSI. These results suggest that the model is not merely a theoretical exercise but a functional tool for developers, capable of executing complex tasks without the typical friction seen in comparable open-weight models.

The emergence of Flint occurs against a backdrop of similar research efforts across the United States. While the extent of progress at other labs remains unclear, published work from teams at Microsoft, Carnegie Mellon University, and the University of Virginia indicates that fully recursive systems are already feasible. Recent studies have demonstrated how an AI coding agent can recursively improve itself by modifying both the software harness that converts reasoning into action and its own model weights. This convergence of research suggests that the capabilities demonstrated by Recursive are part of a broader trend toward automating the AI development lifecycle.

Looking ahead, Recursive believes that the approach used to build Flint could make cutting-edge AI more accessible to smaller teams. Socher argued that this method is the only way a very small team can produce a model that surpasses other models in its class. He expressed hope that this success will demonstrate to the public that recursive self-improvement is a beneficial tool rather than a hazard, ultimately leading to advancements that benefit a wider range of users. As the debate over regulation and safety continues, the practical success of Flint stands as a counterpoint to the theoretical risks, highlighting the tangible gains achievable through automated model development.

The release of Flint underscores a shift in how AI models are constructed, moving away from purely human-led engineering toward systems where AI drives the innovation process. For Midwest readers and industry observers, this represents a significant change in the competitive landscape, where smaller teams can leverage recursive systems to compete with larger organizations. The next few months will likely see further scrutiny of both the model’s capabilities and the regulatory efforts aimed at governing its development, as the industry navigates the balance between rapid progress and long-term safety.