Recursive released Flint on Wednesday, October 21, 2026 — a free coding model that runs on a laptop and was built largely by the company’s own AI research system. Why it matters: Flint is an early public test of recursive self-improvement, the idea that AI can help build better AI, delivered as something any developer can download today.
October 21, 2026 — Flint is Recursive’s first model. It is available free on Hugging Face under the Apache 2.0 license, which allows free use and modification.
What happened
Recursive says its automated research system, the Eureka Machine, helped with nearly every step of Flint’s development, from pre-training to post-training and inference. That let a small team build the model from scratch in under four months. Recursive emerged from stealth in May with $650 million in funding at a $4.65 billion valuation, from backers including GV, Nvidia and AMD Ventures. It is led by CEO Richard Socher, Salesforce’s former chief scientist.
The company is careful about what this is and is not. For now, its version of self-improvement is limited: an AI research assistant overseen by people, which produced a model small enough for a laptop. The longer-term goal is full recursive self-improvement, with humans overseeing the process and able to stop it.
The model itself
Flint is a mixture-of-experts model with 40 billion parameters in total, but only 2.5 billion active for any given task. That design is what makes it light enough for an ordinary laptop. Because it runs on your own computer rather than a cloud service, you can use it offline, keep code private, and avoid per-token usage fees.
In Recursive’s own tests, Flint generated about 131 tokens per second on a MacBook Pro with Apple’s M4 Pro chip — roughly twice the 58 to 69 tokens per second similar models managed on the same machine, including Google’s Gemma 4 and Nvidia’s Nemotron 3.5 Lightning. On Nvidia’s RTX Spark it topped 150 tokens per second. On SWE-Bench Verified, a benchmark for fixing real-world software bugs, Recursive says Flint outperformed other models under 50 billion parameters; on the company’s own average across six tests, only Alibaba’s Qwen3.8 27B scored higher. Those are the company’s reported figures, not independent results.
Recursive also acknowledges the trade-off: Flint needs more guidance than the big paid coding assistants, and the company suggests breaking large tasks into smaller steps. Early hands-on testing was encouraging — Wired’s Will Knight used Flint on a Mac with 48GB of memory to build several small programs and reported that it handled his coding requests every time.
Why it matters
Two things are being tested at once. The first is practical: a capable coding model that is free, local and private removes the cost and data concerns that stop many developers from using AI assistants. The second is the research claim behind it. Recursive says Eureka contributed concrete techniques, including a sparser model shape and low-bit training that shrank Flint by 71% with almost no loss in quality. If each new model really does make the research system better at building the next one, Flint is the first public data point — small, supervised, and worth watching rather than taking on faith.
Frequently asked questions
Is Flint free to use?
Yes. Flint is a free download on Hugging Face under the Apache 2.0 open-source license, which allows free use and modification.
What computer do you need to run Flint?
It is designed to run on an ordinary laptop. Recursive reported about 131 tokens per second on a MacBook Pro with an M4 Pro chip, and Wired tested it on a Mac with 48GB of memory.
Was Flint really built by AI?
Largely, according to Recursive. Its Eureka Machine research system contributed to nearly every part of building Flint, but it worked as an assistant overseen by human researchers — not as a fully autonomous system.
Sources: TechCrunch, Wired (hands-on testing, via TechCrunch)