Recursive released Flint on October 21, describing it as a free coding model developed with substantial help from its own automated research system. The launch is interesting for two reasons: the model itself can be evaluated, and the process of using AI to help build AI models is an example of tools improving their own development workflows.
An independent GitHub evaluation published October 22 reports results on an NVIDIA RTX PRO 6000 workstation. That is not equivalent to a 24 GB Apple-silicon Mac mini, and the model’s total parameter count is large despite its sparse active-parameter design. Do not assume it will run smoothly on your machine just because the headline says laptop.
The practical experiment is to test Flint first on a constrained code-editing task in a disposable repository or a supported hosted demo. Compare actual patch quality and time with the tools you already use. Treat the performance numbers as preliminary until the same task runs on the hardware and workflow you care about.