FreeToken is an Apache-2.0, self-hostable MoE inference engine that runs frontier open-weight models locally by pooling GPU, CPU, and host memory as one tier.
Yes, FreeToken is open source under the Apache-2.0 license.
FreeToken is free to use.
Yes, FreeToken can be self-hosted (the source is available under the Apache-2.0 license).
Edge-native Mixture-of-Experts inference engine that runs frontier open-weight models on consumer hardware by treating GPU, CPU, and host memory as a single bandwidth-adaptive tier — the repository's own framing is "datacenter-scale model serving on your desktop." Ships as a desktop app plus a Python interface, under Apache-2.0.
The widely-quoted figures — a 35B model on an 8GB GPU, 2–4× faster than Ollama on MoE — come from the project's arXiv paper (2608.16157) and are not reproduced in the current GitHub README; they are recorded here as paper/vendor claims pending an independent benchmark, not as README-verified facts.