FlashML Publishes FreeToken to Run Frontier MoE Models on Consumer GPUs

FlashML-org has made FreeToken publicly available as an Apache-2.0 project for running large open-weight Mixture-of-Experts models on personal hardware rather than datacenter GPU clusters. The system is presented as an edge-native serving engine that combines GPUs, CPUs, host memory and interconnect bandwidth so consumer machines can serve models whose full expert pools do not fit in VRAM.