FreeToken targets local MoE inference on consumer hardware
FreeToken is an Apache 2.0-licensed, edge-native Mixture-of-Experts serving engine designed to run large open-weight models on personal and consumer hardware. It combines GPUs, CPUs, host memory and interconnects in an elastic platform with bandwidth-adaptive CPU–GPU co-execution, expert caching, dynamic VRAM reallocation and Anthropic- and OpenAI-compatible APIs. The project supports consumer laptops, gaming desktops and workstation GPUs. Its publisher claims it can run 290B-plus frontier MoE models on a gaming PC at interactive speeds, but the supplied report provides no independent benchmark or verified throughput and latency figures.
FreeToken is designed to extend local MoE inference to large open-weight models by coordinating heterogeneous consumer hardware and exposing compatible APIs. Its headline 290B-plus gaming-PC performance claim remains an unverified vendor claim.
FreeToken brings datacenter-scale model serving to your desktop. Run massive models locally, fast and efficiently.
| Download | Paper | Developer Slack | Community Discord | Community WeChat |
#FreeToken#FlashML-org#DeepSeek-V4-Flash#Qwen3.6-35B-A3B#GLM-5.2#Codex#Claude Code#OpenCode#OpenClaw#DeepSeek Harness