Recursive’s Flint is a free laptop coder and a test of AI-built AI

Recursive, the startup led by former Salesforce chief scientist Richard Socher, has released Flint, a free coding model under the Apache 2.0 license that runs locally, TechCrunch reports. The company says its automated research system, the Eureka Machine, contributed to nearly every stage of building it, letting a small team go from scratch to release in under four months. Recursive emerged from stealth in May with $650 million at a $4.65 billion valuation from backers including GV, Nvidia and AMD Ventures, and signed a $410 million compute deal with AWS in June.

Flint is a mixture-of-experts model that activates about 2.5 billion of its 40 billion parameters at a time. In Recursive’s tests it ran around 131 tokens per second on an M4 Pro MacBook Pro, roughly double similar models on the same machine, and beat other sub-50-billion-parameter models on SWE-Bench Verified. An early hands-on review at WIRED was positive.

A capable coding assistant that keeps code on the user’s machine and charges nothing per token is valuable for developers, students and anyone handling sensitive code. It joins a growing set of small open coding models, such as JetBrains’ Mellum2.1 (Mellum2.1), and tools that swap open models into existing coding agents (Together Link).

Most of the performance figures are Recursive’s own, so independent benchmarks will matter. The company itself says Flint needs more guidance than large paid assistants and suggests breaking tasks into smaller steps, framing it as a strong local tool rather than a replacement for frontier cloud models.

Recursive credits the Eureka Machine with a sparser model shape, low-bit training that cut Flint’s size by 71 percent with little quality loss, and custom kernels. It did not say how much researcher time that saved, so it is clear AI contributed but not how much of the judgment was human, which is central to the debate over automated AI research. The trend is broader: OpenAI said last month that by mid-August its agents were logging 3.1 workdays for every one worked by a human researcher, and it has described recursive training supervision for Astra (Astra launch).

Recursive’s stated long-term goal is full recursive self-improvement, with humans overseeing and able to stop it. Today the system is a supervised research assistant producing a laptop model, a low-risk setting. The company’s framing, that each model should make the system a better researcher and build the next faster, describes a loop in which oversight must scale as quickly as capability. Recursive has not published what its stop mechanisms look like. More in our startups coverage, research coverage and AI tools coverage.