Recursive's Flint Brings AI-Assisted Development to Laptop Coding

Recursive has introduced Flint, a coding model designed to run on a laptop and developed largely with help from AI, the company says. The release is a two-part experiment: Recursive used its own AI system to assist in building the model, then designed Flint to make coding support less dependent on cloud computing.

Flint is Recursive's first model. The company says a small team built it in under four months, with its Eureka Machine assisting across nearly every stage of development. The model is available to download for free under the Apache 2.0 license, giving developers a way to test it locally and inspect its open weights.

What Flint Is and How Its Mixture-of-Experts Design Works

Flint uses a mixture-of-experts, or MoE, architecture. It has 40 billion parameters in total, but activates 2.5 billion for a given task, according to Recursive. Rather than using every parameter for every request, an MoE model routes work to selected parts of the network.

That design is intended to reduce the computing resources needed to run the model. Recursive says the lower active-parameter count helps make Flint suitable for ordinary laptops. It does not mean every laptop will run Flint equally well: performance depends on hardware and setup, and the launch coverage does not provide specific device requirements or comparable speed and coding benchmark results.

The laptop focus matters because local models can let developers work without sending each coding prompt to a hosted service. But the practical value will depend on how Flint performs on a user's hardware and tasks, including whether its responses are useful enough to offset the effort of installing and running it.

How Recursive Says Its Eureka Machine Helped Build the Model

Recursive says its Eureka Machine contributed to nearly every stage of Flint's development. The company presents the system as a way for AI to help with the work of creating AI models, not simply as a coding assistant used for a few isolated tasks. It says that support helped a small team complete the project in less than four months.

That claim connects Flint to Recursive's broader interest in recursive self-improvement: using AI systems to help develop more capable AI systems. The company ultimately wants to apply that approach beyond model-building, including to scientific work. The idea could speed up research, but it also raises questions about how much development can be automated and how people will evaluate and control systems that help shape their successors.

The claim that AI assisted in building Flint should be distinguished from a claim that the model built itself. Recursive's reported account describes AI help across development; it does not establish that the process was fully autonomous.

Where to Get Flint and What Early Impressions Show

Flint is available on Hugging Face under the Apache 2.0 license. Developers can download it and try it on their own coding tasks, including on a laptop. The supplied launch coverage does not report standardized benchmark results, so there is not yet a basis here for comparing its coding performance directly with other models.

WIRED's reporter tested Flint and described it as an impressive open-weight coding model. That is a publication's hands-on assessment, not a substitute for benchmark data or a guarantee that the model will work well for every developer. Local testing can help establish whether its output, speed and resource needs suit a particular workflow.

Flint therefore offers a concrete early test of Recursive's two ideas: AI-assisted model development and coding assistance designed for local use. Download it from Hugging Face under its Apache 2.0 license and try it on your own tasks to see how it performs on your laptop.