San Francisco AI startup Recursive on Wednesday, Oct. 21, released Flint, a free coding model that runs on an ordinary laptop and that the company says was built largely with the help of its own AI research system.

Why it matters: It is an early public test of "recursive self-improvement," the idea that AI can help build better AI, one of the most debated ideas in the industry, TechCrunch reported.

What Flint is

  • A mixture-of-experts model with about 40 billion parameters, of which only about 2.5 billion are active at a time, according to Recursive.
  • Free to download on Hugging Face under the Apache 2.0 open source license.
  • Runs on your own computer, so developers can work offline, keep code private and skip cloud usage fees.

How fast

In Recursive's tests, Flint produced about 131 tokens per second on a MacBook Pro with Apple's M4 Pro chip, roughly twice the 58 to 69 tokens per second of similar models such as Google's Gemma 4, TechCrunch reported. Those figures come from the company.

Recursive also says Flint "may need more guidance" than larger closed coding models.

Built by AI?

The company says its automated research system, the Eureka Machine, helped with nearly every step, letting a small team build the model in under four months. CEO Richard Socher, formerly chief scientist at Salesforce, said its contributions included low-bit training that shrank Flint by 71% with almost no loss in quality.

Recursive came out of stealth in May with $650 million at a $4.65 billion valuation from backers including GV, Nvidia and AMD Ventures.

More Bay Area AI news: SF startup Encord turns Jenga into robot training data.