Recursive launches Flint, a free laptop coding model built largely by AI

Recursive's first model arrives

AI startup Recursive has released its first model, Flint: a free, open-source coding assistant designed to run on an ordinary laptop. The launch is notable not just for the model itself but for how it was made. According to TechCrunch, Recursive says its automated research system, the Eureka Machine, contributed to nearly every stage of the build, letting a small team develop the model from scratch in under four months.

Recursive, founded last year and led by former Salesforce chief scientist Richard Socher, came out of stealth in May with $650 million in funding at a $4.65 billion valuation, with backing from GV, Nvidia and AMD Ventures, TechCrunch reports.

Small enough for a laptop

Flint is a mixture-of-experts model: of its 40 billion parameters, only 2.5 billion are active for any given task. That design keeps it light enough for laptop hardware. TechCrunch contrasts it with Meta's Muse Spark Open, an open model released the week before Flint that packs 816 billion parameters and needs at least a terabyte of memory.

Because it runs locally rather than on a company's cloud servers, developers can use Flint offline, keep their code private and avoid the usage fees that cloud AI services charge. The model is downloadable from Hugging Face under the Apache 2.0 license, which permits free use and modification.

Speed and benchmark results

In Recursive's own tests, Flint generated roughly 131 tokens per second on a MacBook Pro with Apple's M4 Pro chip — about double the 58 to 69 tokens per second that comparable models, including Google's Gemma 4 and Nvidia's Nemotron 3.5 Lightning, managed on the same machine. On Nvidia's RTX Spark, output passed 150 tokens per second.

On SWE-Bench Verified, a benchmark measuring how well AI fixes real-world software bugs, Flint beat other models under 50 billion parameters. Across an average of six tests compiled by Recursive, only Alibaba's Qwen3.8 27B scored higher.

Recursive acknowledges that Flint needs more guidance than the big commercial coding assistants and suggests dividing large tasks into smaller steps. Early reactions were still favorable. Wired's Will Knight, testing Flint on a Mac with 48GB of memory, used it to build several small programs, including a to-do list app and a tool that searches the web for research. He found other small coding models frequently hit errors or lagged behind Claude or Codex, but Flint handled his requests every time.

An AI that helps build AI

The Eureka Machine generates research ideas, runs experiments and uses the outcomes to decide what to try next. According to Recursive's launch video, it was involved from pre-training through post-training and inference. Its concrete contributions include a sparser model shape, low-bit training that shrank Flint by 71% with almost no loss in quality, and custom kernels that accelerate the underlying computations.

Recursive did not say how much researcher time this saved, but it is not alone in leaning on AI for research. TechCrunch notes that OpenAI said last month that by mid-August its AI agents were logging 3.1 workdays for every one worked by a human researcher.

Why it matters

Flint is an early public test of recursive self-improvement — the contested idea that AI can build better AI, which then builds better AI still. Some researchers worry that self-improving systems could slip beyond human control. Recursive's current version is deliberately narrow: a human-supervised research assistant that produced a laptop-sized model rather than an autonomous system.

If the bet pays off, each new model should take on longer stretches of work without human help. The company's $410 million compute deal with Amazon Web Services in June, described to TechCrunch as "less about headcount and more about agent count," points to where it expects that work to come from. Socher told Wired the long-term goal is full recursive self-improvement, with humans overseeing the process and able to stop it — and, in theory, each new model making the Eureka Machine a better researcher for the next one.