Recursive releases Flint, a free coding model built largely with its own AI system
The open-weight model runs on a laptop and needs more guidance than large paid coding assistants.
Recursive, a startup founded last year, has released Flint, a free open-weight coding model that runs locally on a laptop and was built largely by its own AI research system, the Eureka Machine. The small team says the system assisted almost every stage of development, letting them produce the model from scratch in under four months. Flint is a mixture-of-experts design that activates 2.5 billion of its 40 billion parameters per task, and it is distributed on Hugging Face under the Apache 2.0 license. Recursive reports roughly 131 tokens per second on a MacBook Pro with an M4 Pro chip, about double comparable models, and says Flint beat other sub-50-billion-parameter models on SWE-Bench Verified. The company concedes the model needs more guidance than large paid coding assistants, and an early Wired review found it reliable on small coding tasks; the company's longer-term aim is a Eureka Machine that can carry out recursive self-improvement under human oversight.
Why it matters
Flint shows that a small team can build a capable coding model in a few months with heavy AI assistance, and that such a model can run locally on consumer hardware rather than only through paid cloud services. Its benchmark lead is limited to models under 50 billion parameters, and the company itself says it needs more guidance than the large paid assistants, so its practical standing against those tools is still unclear. The company's stated longer-term goal of an AI system that improves itself under human oversight is an aim it has set out, not something Flint demonstrates.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.
Sources
- TechCrunch