Recursive launches Flint, a free coding model that runs on a laptop
Recursive says its automated research system Eureka Machine took part in nearly the whole development of Flint, which can be downloaded free from Hugging Face.
AI startup Recursive released Flint, a free coding model that runs locally on a laptop, according to TechCrunch, which reported the release on October 21.
Recursive said most of Flint's development was carried out by AI. Its automated research system, Eureka Machine, took part in nearly the whole development process, proposing ideas, running experiments and using the results to decide the next attempt. A small team built the model from scratch in under four months.
The system found a training method that cut model size by 71% while keeping quality nearly unchanged. Flint uses 2.5 billion of its 40 billion total parameters per task, and it runs on the user's computer rather than the company's cloud servers. Developers can use it without an internet connection and without sending code outside, and there are no usage-based fees.
The model can be downloaded free from Hugging Face. In an evaluation of real software bug-fixing ability, Flint scored highest among models under 50 billion parameters, according to the company.
Recursive acknowledged that using Flint requires more guidance than large paid coding AI tools and recommends splitting large tasks into small steps.
Company background
Recursive was founded last year and raised $650 million at a $4.65 billion valuation in May 2026, its first public appearance. Richard Socher, a former Salesforce chief scientist, is its CEO, and the company aims for recursive self-improvement, in which AI builds better AI.
Quick answers
How much does Flint cost?
Flint is free and has no usage-based fees. It can be downloaded from Hugging Face.
Does Flint need an internet connection?
No. Flint runs on the user's computer rather than cloud servers, so code is not sent outside.
How large is the Flint model?
It has 40 billion total parameters and uses 2.5 billion of them per task.