Run a 284B MoE model on a gaming desktop with FreeToken
FreeToken is an edge-native MoE serving engine that treats a personal machine as a unified inference platform, not a small GPU. Instead of a fixed offload strategy, it continuously maps computation and model state across CPU, GPU, and host memory as agent workloads shift. That translates to a 35B model on an 8GB laptop, 284B on a gaming desktop, and up to the 753B GLM-5.2 on a single workstation GPU. It installs via pip and exposes Anthropic/OpenAI-compatible APIs for tools like Claude Code.