M1: TOWARDS SCALABLE TEST-TIME COMPUTE WITH MAMBA REASONING MODELS

Summary from the TLDR AI newsletter

M1 is a Mamba reasoning model that has been trained with extended test time computation. It doesn't fully match state-of-the-art models, but it shows strong performance, especially on long context and through-put.