Is Flint the Killer of Qwen 3.8 27B? (First RSI Model)

Recursive Flint writes 131 tokens per second on an M4 MacBook Pro, ten times the 13 tok/s of Qwen 3.8 27B. So is Flint the Qwen 3.8 27B killer, and is it really the first RSI model? Per token, yes. Per finished task, not yet. Flint is Recursive AI’s first model, a 39.8B mixture of experts (MoE) coding model with about 2.5B parameters active per token. It’s open weights under Apache 2.0, with a 262K context, text only. Richard Socher’s lab says its Eureka Machine, an automated AI research system, helped build nearly every part of it in under four months. That is why people are calling it the fi