Inkling-Small is a 276B-total, 12B-active open-weights MoE model that the lab says surpasses Inkling on reasoning and agentic coding.
Thinking Machines Lab published Inkling, a Mixture-of-Experts model with 975B total and 41B active parameters, with full weights on Hugging Face under Apache 2.0.