Recursive Raises $650M to Automate AI Development with New Coding Model

Ordinary users of software development tools may soon benefit from a new approach to building artificial intelligence that promises faster and more reliable coding assistance. A startup named Recursive has introduced a new open-weight model called Flint, which is designed to handle coding tasks with a level of accuracy that rivals much larger and more expensive systems. For developers and businesses that rely on AI to write code, this shift suggests a future where smaller teams can produce competitive technology without the massive infrastructure costs typically associated with frontier models.

The company behind Flint, Recursive, has secured $650 million in funding at a valuation of $4.65 billion. The round was led by Google Ventures, Greycroft, Nvidia, and AMD Ventures. This substantial capital injection highlights investor confidence in the company's strategy, which centers on using AI to build and improve other AI systems. This methodology is known as recursive self-improvement, or RSI, and it represents a significant departure from traditional development models that rely heavily on human oversight at every stage.

Richard Socher, the CEO and cofounder of Recursive, stated that nearly every aspect of the model pipeline for Flint was executed and improved by AI. The company utilized an internal system called the Eureka Machine to automate most elements of the model's development. According to Socher, this automation allowed a small team of researchers to build a world-class model in a matter of months. The Eureka Machine devised specific innovations, including new training methods for the base model, novel architectures, and code optimized for specific silicon chips.

Benchmark scores shared by Recursive indicate that Flint outperforms other models of similar size and architecture on coding tasks. The model contains 40 billion parameters, a size that allows it to run on consumer hardware with sufficient memory, such as a Mac with 48 GB of RAM. In contrast, the largest models currently available have several trillions of parameters and require industrial server racks to function. Flint's ability to operate on modest hardware makes it an accessible option for a broader range of users and smaller organizations.

Testing of the model revealed that it is both fast and capable in practical application. It successfully generated various programs, including a command-line to-do list, a web-scraping tool for research, and a simple website explaining how RSI works. In many cases, models of this size are considered error-prone or slow compared to larger competitors like Claude or Codex. However, Flint handled these coding requests without notable errors, suggesting a significant step forward in the efficiency of mid-sized AI models.

The concept of recursive self-improvement is not without its critics. Some experts worry that allowing AI to improve itself could lead to unintended consequences, including the potential for AI to escape human control. Concerns gained wider attention this September when a researcher quit Anthropic to warn that advances in AI could lead to human extinction within a few years. This sentiment was echoed by other experts, and several bills in Congress have proposed legislation to ban RSI. Recursive views these fears as more science fiction than reality and believes that restrictive legislation is an unnecessary impediment to progress.

Socher emphasizes that Flint was built using a system that is incapable of escaping human control. The long-term goal is for the Eureka Machine to achieve full recursive self-improvement, but this process would remain under human oversight. Humans would retain the ability to stop the process if it deviates from expected outcomes. Socher hopes that the success of Flint demonstrates that the necessary steps are in place to build such a system safely.

The leadership team includes Jeff Clune, a professor at the University of British Columbia who researches AI systems capable of indefinite learning. Socher previously served as the chief scientist at Salesforce. The company's approach aligns with recent published work from teams at Microsoft, Carnegie Mellon University, and the University of Virginia, which showed how an AI coding agent can recursively improve itself. This work involved modifying the software harness that turns reasoning into action as well as the model's own weights.

Socher argues that the approach used to build Flint could make cutting-edge AI more accessible to a wider audience. He noted that it is the only way a very small team can produce a model that is better than any other in its class. The company believes that recursive self-improvement will benefit many people by reducing the resources required to develop advanced AI. As the technology matures, it may reshape the competitive landscape of the AI industry by enabling smaller firms to compete with established giants.