Wire flash
Recursive releases Flint, a 40B open-weights coding model built by its Eureka Machine
Editorial responsibility
- No named human review is recorded for this page.
- Source reporting is collected, normalized, translated or condensed automatically when needed.
- Automatically published source-backed update
Recursive released Flint, its first model, a 40B mixture-of-experts coding AI with open weights under Apache 2.0, built in under four months with help from an AI research system called the Eureka Machine. The Eureka Machine runs the scientific method in a loop, pursuing many research paths in parallel and improving its research from results. Its work appeared in architecture, low-bit training, specialized inference kernels, and speculative decoding. Flint runs entirely on a MacBook at 125+ tokens a second at short contexts, with no per-token bill. The author says this is the closest local coding has felt to viable, since an agent can run all day without a meter and prompts and code stay on owned hardware, and calls it one of the first real milestones in self-improving AI.
Source report
An AI helped build a new coding AI in under four months — and the result runs on a MacBook at over 125 tokens per second at short contexts. It runs entirely on your own machine, so there's no bill per token.
Flint: Recursive's First Release
That model is Flint, the first release from Recursive. It's a 40B mixture-of-experts model with open weights under Apache 2.0.
The research system behind it, the Eureka Machine, runs the scientific method in a loop — chasing many research paths in parallel and improving its own research from whatever results come back.
Where the Research Showed Up
The system's work appeared across several areas:
- Architecture
- Low-bit training
- Specialized inference kernels
- Speculative decoding
If you set out to make a 40B model fly on a laptop, that's the to-do list you'd write. The mixture-of-experts design only wakes up a slice of the model for each token, and the research system targeted the levers that turn that into real speed. So when you watch it run on a MacBook, you're watching the research system's work.
Why Local Matters
Running models locally has long been a goal for developers, and this is the closest it has felt. When the model lives on your machine, you can leave an agent running all day without a meter ticking in the background — and your prompts and code stay on hardware you own.
A Milestone for Self-Improving AI
There's been a great deal of hype around AI that improves itself. This may be remembered as one of the first real milestones: a small team and their research system shipped a model anyone can download and run on their own laptop.
Source
aakashguptaNeutral / independent
Part of this Story
Recursive releases Flint, a 40B open-weight coding model built largely by its Eureka Machine