Recursive, an AI startup founded last year, emerged from stealth in May with $650 million in funding at a $4.65 billion valuation, from backers including GV, Nvidia, and AMD Ventures. Led by CEO Richard Socher, Salesforce’s former chief scientist, the company released its first model, Flint, on Wednesday. Flint is a free coding model that can run on a laptop and was built largely with the help of AI. Recursive’s automated research system, called the Eureka Machine, helped with nearly every step of Flint’s development, allowing a small team to build the model from scratch in under four months. The release is an early public test of recursive self-improvement, the idea that AI can build better AI, which in turn builds even better AI. That concept is among the industry’s most debated; some researchers fear AI improving itself could slip out of human control, but Recursive’s current version is a limited, human-overseen research assistant that produced a laptop-sized model.

Because Flint runs on a user’s own computer rather than a company’s cloud servers, developers can use it offline, keep their code private, and avoid usage fees charged by AI cloud services. The model is free to download on Hugging Face under the Apache 2.0 open-source license. Flint is a mixture-of-experts model, activating only 2.5 billion of its 40 billion parameters for any given task, making it light enough for an ordinary laptop. That contrasts sharply with Meta’s Muse Spark Open, an open model released last week that uses 816 billion parameters and needs at least a terabyte of memory.

Recursive says Flint is also fast. In its tests, Flint generated about 131 tokens per second on a MacBook Pro with Apple’s M4 Pro chip, roughly twice the 58–69 tokens per second that similar models like Google’s Gemma 4 and Nvidia’s Nemotron 3.5 Lightning managed on the same machine. On Nvidia’s RTX Spark, Flint topped 150 tokens per second. On SWE-Bench Verified, which measures how well AI fixes real-world software bugs, Flint outperformed other models under 50 billion parameters. On Recursive’s own average across six tests, only Alibaba’s Qwen3.8 27B scored higher. The company acknowledged that Flint needs more guidance than big paid coding assistants and suggested breaking large tasks into smaller steps. Early reviews were positive: Wired’s Will Knight used Flint on a Mac with 48GB of memory to build several small programs, including a to-do list app and a tool that scours the web for research, and found that while smaller coding models often ran into errors or were slow, Flint nailed his coding requests every time.

The Eureka Machine works by coming up with ideas, running experiments, and using the results to decide what to try next. According to Recursive’s launch video, it contributed to nearly every part of building Flint, from pre-training to post-training and inference. Its contributions include a sparser model shape, low-bit training that shrank Flint by 71% with almost no loss in quality, and custom kernels that speed up computations. Recursive did not say how much time the Eureka Machine saved, but the company previously told TechCrunch that its $410 million compute deal with AWS in June was “less about headcount and more about agent count.” Socher told Wired that the longer-term goal is for the Eureka Machine to achieve full recursive self-improvement, with humans overseeing the process and able to stop it if something goes wrong. Each new model, Recursive says, should make the Eureka Machine a better researcher and help build the next one faster.