FreeToken is an edge-native mixture-of-experts serving engine for running frontier-scale open-weight models on personal hardware.
Follow FreeToken to get an email when its listing changes, such as new pricing or features. One digest a day at most, and you can unsubscribe anytime.
It treats GPUs, CPUs, host memory and interconnects as one elastic inference platform and exposes Anthropic- and OpenAI-compatible APIs for coding and tool-calling agents.
- 01
Bandwidth-adaptive CPU-GPU co-execution with double-buffered prefill streaming and global LRU expert caching
- 02
Semantic anchor checkpoints so agentic context edits avoid recomputing KV and recurrent state
- 03
Runtime VRAM rebalancing between expert cache and KV memory without restarts
- 04
Support for DeepSeek-V4-Flash, Qwen3.6, GLM-5.2, gpt-oss and GGUF checkpoints in MXFP4, NVFP4, FP8 and BF16
- 05
Anthropic- and OpenAI-compatible APIs for Codex, Claude Code, OpenCode and OpenClaw, via a desktop app or uv/pip CLI
Free
FreeToken is free and open-source under the Apache-2.0 license, including the Windows and Linux desktop app. There are no paid tiers; you supply your own NVIDIA RTX 30, 40 or 50 series GPU and host memory.
Best for
Developers with an NVIDIA gaming or workstation GPU who want to run 100B-300B-class open MoE models locally as a fast backend for coding agents.
Not ideal for
Apple Silicon, AMD or CPU-only users, and anyone who needs a mature, long-stable serving stack, since the project targets NVIDIA RTX hardware and is at version 0.2.
A research-backed, fast-moving local MoE server that brings datacenter-style serving techniques to consumer NVIDIA GPUs and is a strong local backend for agent workloads.
v0.2.0 (October 21, 2026) and v0.2.1 (October 26, 2026) added image input from llama.cpp mmproj files, DeepSeek-V4 and Qwen3-VL GGUF architectures, Flint support and host-RAM embedding tables; a November 2, 2026 commit added gpt-oss GGUF serving. The accompanying arXiv paper (2608.16157) describes the design.
As of
Is FreeToken free?+
Yes - FreeToken is Free. FreeToken is free and open-source under the Apache-2.0 license, including the Windows and Linux desktop app. There are no paid tiers; you supply your own NVIDIA RTX 30, 40 or 50 series GPU and host memory.
Is FreeToken open source?+
Yes - FreeToken is open source. FreeToken is free and open-source under the Apache-2.0 license, including the Windows and Linux desktop app. There are no paid tiers; you supply your own NVIDIA RTX 30, 40 or 50 series GPU and host memory.
Who is FreeToken best for?+
Developers with an NVIDIA gaming or workstation GPU who want to run 100B-300B-class open MoE models locally as a fast backend for coding agents.
Who is FreeToken not ideal for?+
Apple Silicon, AMD or CPU-only users, and anyone who needs a mature, long-stable serving stack, since the project targets NVIDIA RTX hardware and is at version 0.2.
What are the best FreeToken alternatives?+
The closest FreeToken alternatives on ai.dosa.dev are Copperhead, NVIDIA Personal AI Router (PAIR), WebLLM - all listed under Hardware & Edge AI.
Who makes FreeToken?+
FreeToken is developed by FlashML. It is listed in the Hardware & Edge AI category on ai.dosa.dev.
Favorite this tool to revisit it later, or Zap it to contribute to the public vote count.