Startup Recursive launches Flint, a laptop coding model largely built with AI
US AI startup Recursive on Wednesday released Flint, a free coding model that runs on a laptop and was largely developed with help from the company's own AI research system. It is an early public test of the idea that AI can improve itself.
Recursive, an AI startup founded last year, on Wednesday released its first model, Flint. It is a free coding model that can run on an ordinary laptop. The company says its automated research system, the Eureka Machine, helped with nearly every step of development, allowing a small team to build the model from scratch in under four months.
Flint is an early public test of recursive self-improvement: the idea that AI can build better AI, which then builds better AI still. It is among the most debated ideas in the industry, and some researchers fear self-improving AI could slip out of human control. For now, Recursive's version is narrower: a human-supervised AI research assistant.
The company
Recursive came out of stealth in May with $650 million in funding at a $4.65 billion valuation. Backers include GV, Nvidia and AMD Ventures. The CEO is Richard Socher, formerly chief scientist at Salesforce.
How the model works
Because Flint runs on the user's own machine rather than cloud servers, developers can work offline, keep code private and avoid usage fees charged by cloud AI services. It can be downloaded free from Hugging Face under the Apache 2.0 open-source license.
Flint is a mixture-of-experts model, using only 2.5 billion of its 40 billion parameters at a time. By comparison, Meta's Muse Spark Open, released last week, has 816 billion parameters and needs at least a terabyte of memory.
In Recursive's tests on a MacBook Pro with an Apple M4 Pro chip, Flint produced about 131 tokens per second, against 58 to 69 for similar models such as Google's Gemma 4 and Nvidia's Nemotron 3.5 Lightning. On SWE-Bench Verified, a benchmark of fixing real-world software bugs, it outperformed other models under 50 billion parameters. On the company's own six-test average, only Alibaba's Qwen3.8 27B scored higher.
Recursive acknowledges Flint needs more guidance than large paid coding assistants and suggests splitting big tasks into smaller steps. Wired's Will Knight reported that it handled his coding requests every time.
The Eureka Machine
The system proposes ideas, runs experiments and uses results to choose what to try next. Recursive says it contributed to pre-training, post-training and inference, including low-bit training that shrank Flint by 71% with almost no loss in quality. The long-term goal is full recursive self-improvement, with humans overseeing the process.
