Recursive, an AI startup founded last year, has released its first model, a free coding tool called Flint that can run on a laptop. The company says the model was built largely with the help of AI, and the launch took place on Wednesday, October 21, 2026.

According to Recursive, an automated research system it calls the Eureka Machine contributed to nearly every stage of Flint's development. The company says that allowed a small team to build the model from scratch in less than four months. The system proposes ideas, runs experiments and uses the outcomes to choose what to try next. Recursive's launch video says it took part in pretraining, post-training and inference work.

Flint serves as an early public test of recursive self-improvement, the idea that AI can build better AI, which then builds better AI still. The concept is among the most contested in the industry, and some researchers worry that self-improving AI could slip beyond human control. Recursive's current approach is narrower: a research assistant supervised by people that produced a model compact enough for a laptop.

Because Flint runs on a user's own machine rather than on a company's cloud servers, developers can work offline, keep their code private and avoid usage fees charged by cloud AI services based on the amount of text processed. The model can be downloaded for free from Hugging Face under the Apache 2.0 license, which permits free use and modification.

Flint uses a mixture-of-experts design, activating only 2.5 billion of its 40 billion parameters for any given task. That makes it far smaller than Meta's Muse Spark Open, an open model for Macs released last week, which has 816 billion parameters and needs at least a terabyte of memory.

In Recursive's tests, Flint generated about 131 tokens per second on a MacBook Pro with Apple's M4 Pro chip. Comparable models, including Google's Gemma 4 and Nvidia's Nemotron 3.5 Lightning, managed 58 to 69 tokens per second on the same machine. On Nvidia's RTX Spark, Flint exceeded 150 tokens per second. On SWE-Bench Verified, which measures how well AI fixes real-world software bugs, it outperformed other models under 50 billion parameters. Across six tests averaged by Recursive, only Alibaba's Qwen3.8 27B scored higher.

The company acknowledged that Flint needs more guidance than the large paid coding assistants and recommends splitting big tasks into smaller steps. Early reviews were nonetheless favorable. Wired's Will Knight used it on a Mac with 48GB of memory to build programs including a to-do list app and a web research tool, and found that Flint handled his requests every time, whereas smaller coding models often hit errors or ran slowly compared with Claude or Codex.

CEO Richard Socher, Salesforce's former chief scientist, said on social media that Eureka's contributions included a sparser model shape, low-bit training that shrank Flint by 71% with almost no loss in quality, and custom kernels, which are specialized programs that speed up core computations. Recursive did not say how much researcher time the system saved. It is not alone in relying on AI for research: OpenAI said last month that by mid-August its AI agents were logging 3.1 workdays for each one worked by a human researcher.

Recursive emerged from stealth in May with $650 million in funding at a $4.65 billion valuation, backed by investors including GV, Nvidia and AMD Ventures. It also struck a $410 million compute deal with Amazon Web Services in June, which it described as "less about headcount and more about agent count." Socher told Wired the long-term aim is for the Eureka Machine to reach full recursive self-improvement, with humans overseeing it and able to halt it if something goes wrong.