Recursive Launches Flint, A Laptop Coding Model Built With AI
Recursive says its free coding model runs locally and was developed largely with help from its AI research system, the Eureka Machine.
Recursive has released Flint, its first AI model, a free coding tool designed to run on a laptop. The company says its Eureka Machine research system helped develop nearly every part of the model, allowing a small team to build it in under four months, according to TechCrunch. Flint offers a public test of Recursive’s ambition to use AI to improve AI, though people still oversee the process.
Table of Contents
A coding model that runs locally
Flint is available to download from Hugging Face under the Apache 2.0 license, which permits free use and modification, TechCrunch reported. Running it on a user’s computer rather than a cloud service means developers can work offline, keep code on their own machines and avoid fees tied to cloud-model usage.
The model has 40 billion parameters, but its mixture-of-experts design uses 2.5 billion at a time. Recursive says that helps make it practical to run on a laptop. The company also cautions that Flint may need more guidance than large paid coding assistants and recommends splitting bigger jobs into smaller steps, according to TechCrunch.
Performance claims and an early test
In Recursive’s tests on a MacBook Pro with an M4 Pro chip, Flint generated about 131 tokens per second, versus 58 to 69 for several comparable models on the same machine, TechCrunch reported. The outlet also said Flint led models under 50 billion parameters on the SWE-Bench Verified software-bug benchmark. These are company-reported results, not a broad independent assessment.
Wired’s Will Knight tested Flint on a Mac with 48 GB of memory and found it fast and capable when building small programs, including a command-line to-do list and a research-gathering tool. His experience was positive, but it was a hands-on test rather than evidence of how Flint will perform across every coding task.
How AI helped build the model
Recursive says the Eureka Machine proposes ideas, runs experiments and uses the results to choose further work. TechCrunch reported that the system contributed across Flint’s development, including training, model design and specialized code to speed computations. CEO Richard Socher told Wired that a handful of researchers built the model in a few months with that assistance. Recursive has not said how much researcher time the system saved, TechCrunch noted.
The release illustrates two possible uses for developers: a coding assistant that keeps work local, and AI-assisted research that helps a small team build a model. It does not establish that AI can improve itself without people. Socher told Wired that Recursive hopes the Eureka Machine will eventually achieve full recursive self-improvement, while remaining subject to human oversight and a human decision to stop it if necessary. Wired also reported that the prospect of such systems has prompted safety concerns and proposed restrictions in Congress.
Sources
This story was compiled by AI from the reports below. Read the originals for the full details.