Meet FreeToken: An Edge-Native MoE Serving Engine that Runs 753B GLM-5.2 on a Single Workstation GPU

FreeToken is a new serving engine designed to run large Mixture-of-Experts models locally by balancing data between a workstation GPU and CPU memory. By optimizing how cache misses are handled, the engine enables the execution of the 753B GLM-5.2 model on consumer-grade hardware.