Recursive says AI built most of its new laptop coding model

Of Flint’s 40 billion parameters (the settings a model learns), only 2.5 billion are active for each token, a chunk of text, it produces. On an M4 Pro MacBook Pro, Recursive clocks it at more than 125 tokens a second. Running locally means no per-use bill, and your code never leaves your computer.

The company says a small team built Flint from scratch in under four months with its Eureka Machine, software that does the research loop itself: it comes up with an idea, tries it and lets the result steer the next attempt. It shaped the model’s design, training and the code that runs it on Apple and Nvidia chips.

AI improving AI alarms some researchers, enough that several bills before Congress would outlaw it, Wired reports. Chief executive Richard Socher, formerly Salesforce’s chief scientist, says humans supervise the system and can halt it at any point. Recursive raised $650 million at a $4.65 billion valuation in May, TechCrunch reports.

Flint brings a strong coding model to an ordinary laptop, with no cloud account needed.