Recursive, a startup led by former Salesforce chief scientist Richard Socher, has shipped its first model. Flint is built for coding, costs nothing and runs on an ordinary laptop with no cloud involved.

The interesting part is how it was made. The company says its automated research system, the Eureka Machine, took part in almost every stage of the work, which let a small team build the model from scratch in under four months.

What Flint can do

It is a mixture-of-experts model: of its 40 billion parameters, only about 2.5 billion are active for any request. That keeps it light enough for a laptop, so code stays on your machine and there are no per-token fees. The weights are on Hugging Face under the Apache 2.0 license.

  • about 131 tokens per second on a MacBook Pro with M4 Pro in Recursive's tests, roughly twice as fast as similar models;
  • beats other models under 50 billion parameters on SWE-Bench Verified;
  • on the company's six-test average, only Qwen3.8 27B scored higher.

AI that builds AI

The Eureka Machine proposes ideas, runs experiments and uses the results to pick the next step. Recursive credits it with a sparser architecture, low-bit training that cut the model size by 71% with almost no quality loss, and custom compute kernels. People still oversee the process.

Recursive admits Flint needs more guidance than the big paid assistants and suggests splitting large tasks into smaller steps.

Our take

For anyone building sites and bots, this is another step toward a useful local assistant: your code stays private and the work doesn't depend on rate limits or a connection. The numbers so far come from the company itself, so test it on your own tasks first. More in the TechCrunch report.