Recursive launches Flint, a fast open-weight coding model built to run locally

Recursive has released its first model, and it is not asking for your API key. Flint is an open-weight coding model designed to run on your own machine, and the company says it pushes past 125 tokens per second on a MacBook Pro. What Recursive actually shipped Flint launched on October 21, 2026. It is a 40-billion-parameter model built specifically for coding work. On a MacBook Pro with an M4 Pro chip, Flint generates over 125 tokens per second. On RTX Spark hardware, Flint could potentially reach 150 tokens per second or more. Recursive’s performance claims say the model can beat comparable models by up to two times. For agentic coding, where the model takes multi-step actions rather than simply autocompleting a line, Flint is described as workable on consumer machines with 48GB of RAM. Flint is also positioned among the leading Mixture of Experts models under 50 billion parameters, particularly for agentic coding. A Mixture of Experts design works a bit like a hospital triage desk: instead of every specialist seeing every patient, the system routes each request to the parts of the model best suited to handle it. The payoff is typically faster responses without running the whole n...