Flint 40B: Frontier Local LLM (Fast 125 Tps on MacBook) (twitter.com)

Flint 40B has been unveiled as Recursive SI's first local large language model (LLM), delivering impressive performance with speeds exceeding 125 tokens per second on a MacBook Pro and 150 tokens per second on an RTX Spark. Developed in just under four months using their innovative Eureka Machine research system, Flint stands out in the AI/ML field for its exceptional coding capabilities and performance efficiency, boasting a speed that is up to 2x faster than its competitors. The significance of Flint lies in its promise of providing private, unmetered access to powerful AI tools directly on users' machines, thereby addressing concerns about data privacy and model performance. With open weights available on Hugging Face, developers can run Flint locally using their preferred frameworks or easily test it through platforms like Opencode or Pidotdev. This marks a notable advancement in local AI solutions, enabling greater accessibility and fostering a growing ecosystem of fast, reliable coding assistants that empower individual developers.