Recursive's Flint is a laptop coding model it says AI mostly built

Recursive, a startup founded in 2025, released Flint, a free coding model small enough to run on a laptop, and its pitch is as much about how it was made as what it does. The company says its automated research system, the Eureka Machine, helped with nearly every step, letting a small team build the model from scratch in under four months with human researchers overseeing the process, as TechCrunch reported. That system also produced the model's sparser design, a low-bit training method that shrank it by 71% with almost no quality loss, and custom kernels.

Flint is a mixture-of-experts model: 40 billion parameters in total, but only 2.5 billion active for any given task, which is what lets it run locally and fast. On a MacBook Pro with an M4 Pro chip Recursive measured about 131 tokens per second, roughly double comparable models, and over 150 on Nvidia's RTX Spark. On SWE-Bench Verified it beat other models under 50 billion parameters, and across Recursive's six-test average only Alibaba's Qwen3.8 27B scored higher. The company is open about the tradeoff: Flint needs more guidance than large paid assistants, so it suggests breaking big tasks into smaller steps, and a Wired reviewer found it reliable on small projects.

Flint is on Hugging Face under the Apache 2.0 license, so it runs offline, keeps code on your own machine, and avoids cloud usage fees.

Why it matters

If you want a coding model that runs on your own laptop without sending code to a cloud, Flint is a usable option today under a permissive license, with the caveat that you will need to feed it smaller, clearer tasks. The claim that AI did most of the building is also an early data point on whether automated research can compress model development from years into months.