FreeToken Open-Sources Datacenter-Scale MoE Inference for Desktops

Researchers from Berkeley and MIT open-sourced FreeToken, a native on-device MoE inference engine that treats GPU, CPU and memory as one elastic platform, letting a gaming PC run a 284B model locally and an 8GB laptop GPU run 35B, quickly topping 15k GitHub stars.

2026-11-01 ~ 2026-11-03 · 2 related posts

• FreeToken brings datacenter-scale MoE model serving to desktops, hits 15k stars — FlashML-org · 2026-11-01
• FreeToken open-sources engine running 284B models on a gaming PC — alex_verem · 2026-11-03