AI startup Recursive unveiled Flint, a free coding model designed to run on laptops, marking a significant milestone in AI research. Founded just last year, Recursive created Flint largely with the help of its own automated research system called the Eureka Machine. This system played a critical role throughout Flint’s development, enabling the startup’s small team to build the model from scratch in under four months. Flint represents an early trial of recursive self-improvement, an approach where AI can iteratively improve itself with human oversight.

Recursive emerged from stealth mode in May 2026 after raising $650 million at a $4.65 billion valuation, backed by notable investors such as GV, Nvidia, and AMD Ventures. CEO Richard Socher, formerly Salesforce’s chief scientist, leads the company’s push to democratize AI tools by providing Flint as an open source project available on Hugging Face under the Apache 2.0 license. Flint’s design enables it to run efficiently on personal laptops, offering offline access, privacy of code, and freedom from expensive cloud service fees based on usage.

Technically, Flint is a mixture-of-experts model, selectively using portions of its 40 billion parameters—about 2.5 billion at a time—which keeps its operation lightweight enough for typical laptop hardware. Performance tests showed it generating code tokens at speeds nearly double those of comparable models on a MacBook Pro with Apple's M4 Pro chip. Flint’s coding accuracy also stood out in benchmarks aimed at fixing software bugs, outclassing most models below 50 billion parameters and receiving commendations for its reliability in coding tasks.

The Eureka Machine’s contributions extended beyond coding assistance; it refined Flint’s architecture, optimized training with techniques like low-bit quantization that shrunk model size by 71%, and developed custom computational kernels to boost speed. Recursive aims to gradually reduce human intervention as this AI-driven research engine improves itself, with the long-term goal of achieving fully autonomous self-improvement while ensuring safe human supervision. The company’s recent $410 million compute deal with Amazon Web Services underscores its ambition to scale AI “agent count” rather than just researcher headcount in future projects.