Recursive Startup Launches Flint 40B Model

Recursive’s new AI model Flint writes code so well that it beats larger rivals with just 40 billion parameters. The system, built by a handful of researchers, was assembled by the company’s own Eureka Machine. CEO Richard Socher says nearly every step of the model pipeline was created and refined by AI. Flint’s benchmark scores on coding tasks outrank other 40‑billion‑parameter models, according to Recursive’s data. The startup raised $650 million, valuing it at $4.65 billion, with investors like Google Ventures and Nvidia. Critics warn that endless recursive self‑improvement could slip beyond human control, and several US bills now seek to ban it. Socher counters that the current system cannot escape oversight and that they have “all the steps in place”. If the Eureka Machine ever achieves full recursion, humans will still be able to halt the process, he assures.