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

Recursive, a startup valued at $4.65 billion, has released Flint, a 40-billion-parameter open-weight model that demonstrates significant progress toward recursive self-improvement. The model is the first product from a company that relies heavily on artificial intelligence to design and train its own software, a strategy that allows a small research team to compete with much larger competitors. Flint has shown strong performance in coding benchmarks, outperforming other models of comparable size and architecture.

The company’s approach centers on a system called the Eureka Machine, which automated nearly every aspect of Flint’s development pipeline. According to CEO and co-founder Richard Socher, the Eureka Machine devised new training methods, model architectures, and code optimized for specific silicon chips. This automation enabled Recursive to build a world-class model in just a few months with a handful of researchers, a timeline that would typically require significantly more human resources at other firms.

Flint’s capabilities were tested on a Mac with 48 GB of RAM, a setup sufficient for a model of its 40-billion-parameter scale. In practical use, the model proved fast and reliable, successfully generating a command-line to-do list, a web-scraping tool, and a simple website explaining recursive self-improvement. These results contrast with other models of similar size, which are often slower or more error-prone compared to larger, more expensive systems like Claude or Codex.

The broader strategy behind Flint is recursive self-improvement, or RSI, a process where AI systems contribute to their own refinement. While AI is already used to assist in model development, RSI aims to create a loop where the system identifies and implements its own improvements. If fully realized, this could allow for indefinite advancement of AI capabilities with minimal human intervention, potentially making cutting-edge AI more accessible to organizations with limited resources.

However, RSI remains a controversial concept in the tech industry. Some experts worry that allowing AI to improve itself without sufficient human oversight could lead to unforeseen consequences, including the potential for systems to escape control. These concerns gained prominence this September when a researcher from Anthropic warned that advances in RSI could lead to human extinction within a few years. The debate has even reached the legislative level, with several bills in Congress proposing to ban recursive self-improvement.

Recursive views these fears as overly dramatic, comparing them to science fiction rather than imminent reality. The company argues that legislation to prevent RSI is an unnecessary barrier to progress. Socher emphasizes that Flint was built using a system that remains under human control. The long-term goal is for the Eureka Machine to achieve full RSI, but the process would be overseen by humans who retain the ability to stop it if necessary.

The company’s financial backing reflects investor confidence in this approach. Recursive has raised $650 million from Google Ventures, Greycroft, Nvidia, and AMD Ventures. The team includes Socher, formerly the chief scientist at Salesforce, and co-founder Jeff Clune, a professor at the University of British Columbia who researches AI systems capable of indefinite learning.

While the full extent of RSI’s risks remains unclear, recent published work from Microsoft, Carnegie Mellon University, and the University of Virginia suggests that fully recursive systems are feasible. These findings support the idea that AI coding agents can improve themselves by modifying their own software harness and model weights. For now, RSI appears to be a promising method for enhancing AI models, potentially allowing smaller teams to produce superior technology.