Thinking Machines Lab Releases Inkling-Small, a 276B-Total Open MoE Model

Inkling-Small is a Mixture-of-Experts model with 276B total parameters and 12B active, trained on NVIDIA GB300 NVL72 systems. It is about a quarter the size of Thinking Machines' Inkling, which has 975B total and 41B active parameters.