M1: The Next Step in Test-Time Computing
M1 is a new Mamba reasoning model, trained to handle extended test-time computation. While it doesn’t quite match the cutting-edge models, it’s making a strong case for itself, especially in areas like long context processing and throughput. Imagine it as a solid, reliable performer, even if it’s not quite at the top tier yet. It’s a big step forward for scalable test-time compute.