A startup called Recursive has released an open-weight AI coding model named Flint that was developed largely through recursive self-improvement, a technique where AI systems autonomously refine and enhance AI models, according to a report published in Wired. The company used its proprietary recursive AI system, called the Eureka Machine, to automate nearly every element of Flint's development pipeline. This allowed a small team of researchers to create a world-class model within just a few months, demonstrating that fully recursive self-improvement is becoming increasingly practical.
Flint contains 40 billion parameters and can run on consumer hardware such as a Mac with 48 GB of RAM, unlike the largest models that require trillions of parameters and industrial server racks. Benchmark scores shared by Recursive indicate that Flint surpasses other models of comparable size and architecture on coding tasks. The Eureka Machine generated multiple innovations during Flint's creation, including a novel approach to training the base model, fresh model architectures, and code optimized for particular silicon chips. Testing showed the model to be both fast and highly capable, completing coding requests accurately without the errors or slowness typical of models in its size class.
Recursive CEO and cofounder Richard Socher told Wired that "nearly every aspect of the entire model pipeline was actually done and improved upon by AI." Socher said the approach used to build Flint could democratize cutting-edge AI, noting it's "the only way this very small team can produce a model that's actually better than any other model in its class." He added that the company hopes to show "that indeed we have all the steps in place" to build a fully recursive self-improvement system, though one that would remain under human oversight and could be halted if problems emerge. The report notes that Recursive sees concerns about AI escaping human control as "more sci-fi than reality" and views proposed legislation to ban recursive self-improvement as an unnecessary obstacle to progress.
Recursive self-improvement has sparked intense debate because some experts fear that AI capable of improving itself indefinitely could spiral beyond human control and potentially threaten humanity. These worries gained prominence in September when a researcher left Anthropic and cautioned that advances in AI, including recursive self-improvement, might lead to human extinction within years—a warning subsequently echoed by numerous other experts. The controversy has grown significant enough that several congressional bills now propose banning the technique. Yet published research suggests fully recursive systems are already achievable—a team from Microsoft, Carnegie Mellon University, and the University of Virginia recently demonstrated how an AI coding agent can recursively improve itself by modifying both the software framework that converts reasoning into action and its own model weights.
Recursive has secured $650 million in funding at a $4.65 billion valuation from investors including Google Ventures, Greycroft, Nvidia, and AMD Ventures. Socher previously served as chief scientist at Salesforce, while his cofounders include Jeff Clune, a University of British Columbia professor researching AI systems capable of learning indefinitely. The company aims to use recursive self-improvement to advance not only AI itself but also other scientific fields through AI applications. Socher told Wired that while the goal is for the Eureka Machine to eventually achieve full recursive self-improvement, human operators would maintain oversight throughout, with the ability to stop the process if it veers off course. The business case for smaller teams leveraging automation to compete with larger labs may reshape competitive dynamics across the sector, though the regulatory landscape remains uncertain as policymakers weigh innovation against speculative risks.