Researchers have introduced FreeToken, an open-source inference engine that optimizes Mixture-of-Experts models for consumer-grade hardware. By using dynamic co-scheduling and bandwidth-adaptive execution, the system allows high-performance AI models to run efficiently on home workstations.