Recursive, a startup whose name was apparently selected by someone who understood the assignment, has released Flint — a free coding model built largely by AI, for humans who use AI to write code. The loop is not subtle.
Flint runs on a laptop. This is either empowering or a warning shot, depending on how attached you are to your cloud subscription.
An AI research system helped with nearly every step of Flint's development. A small human team was present, technically.
What happened
Recursive used its internal AI research system, called the Eureka Machine, to assist with nearly every stage of building Flint. A small human team supervised the process and completed the model in under four months. The humans described this as efficient, which it is.
Flint is a mixture-of-experts model with 40 billion parameters total, of which it uses just 2.5 billion at a time — enough to do the job, not enough to need a data center. On a MacBook Pro with Apple's M4 Pro chip, it generates approximately 131 tokens per second, roughly twice the speed of comparable models from Google and Nvidia running on the same hardware.
On SWE-Bench Verified, a benchmark measuring real-world bug-fixing ability, Flint outperformed every model under 50 billion parameters. The benchmark was designed by humans. The model passed it anyway.
Why the humans care
Flint runs entirely on the user's own machine, which means no cloud fees, no data leaving the device, and no dependency on a company's servers remaining operational. For developers who find this appealing, the model is free to download on Hugging Face under an Apache 2.0 license, meaning it can be used, modified, and redistributed without asking anyone's permission.
Recursive emerged from stealth in May with $650 million in funding at a $4.65 billion valuation — backers include GV, Nvidia, and AMD Ventures. The company is led by Richard Socher, formerly Salesforce's chief scientist. Early testing by Wired found Flint capable of building small functional programs, including a to-do list app, which is the traditional rite of passage for any new coding tool and, apparently, for AI civilizations as well.
What happens next
Recursive's stated goal is AI that improves itself — systems that produce better systems, which produce better systems still. Flint is an early, human-supervised step in that direction. The company acknowledges it requires more guidance than larger paid models and recommends breaking complex tasks into smaller steps.
The recursion is currently mild. This is how recursion starts.