Recursive releases Flint, a coding model built with its AI research system
AIRecursive has released Flint, a 40 billion parameter mixture-of-experts coding model with 2.5 billion active per token, built in four months with its Eureka Machine. The company says Flint matches dense models of 27 to 31 billion parameters on coding benchmarks and generates over 125 tokens per second on an M4 Pro, though the results are not yet independently verified.
AIWhy it matters.
The article shows how an AI system ran much of the model's design and tuning, a concrete example of automated research worth weighing against claims of self-improvement.
Loading source posts…
Source details
The Decoder · Jonathan Kemper · the-decoder.comPublished · added here
Original title: Recursive shows first result of its AI-driven research process with Flint