Serves large Mixture-of-Experts models such as DeepSeek-V4-Flash and GLM-5.2 on consumer NVIDIA GPUs, sharing work across GPU and CPU behind OpenAI- and Anthropic-compatible APIs.
Use it now
Keep it with up to 9 other tools you're setting up, then share them all as one link.
Try this first
Ask your coding agent:
- Coding Agent Launcher
Use FreeToken to discover the served model: ‹what you're looking for›
- MoE Expert Offload
Use FreeToken to run MoE layers: ‹what you need›
Install the CLI
Install it as its own instructions say.
Its page: github.com/FlashML-org/FreeToken
Call the API
Auth None needed
Its page: github.com/FlashML-org/FreeToken
Compare
Press Compare to add it to the tray at the bottom of the window, then pick up to 2 others anywhere on the site.
- Access
- CLIAPI
- Auth
- None
- Skills
- 3
Skills
Compatible API Server
Starts a local server with ft serve that exposes OpenAI /v1 endpoints, the Anthropic Messages API and the Responses API for a chosen model.
Coding Agent Launcher
Discovers the served model with ft launch, writes the coding agent's provider config, installs its CLI if missing and launches it.
MoE Expert Offload
Runs MoE layers with bandwidth-adaptive CPU-GPU co-execution and LRU expert caching, reallocating VRAM between experts and KV at runtime.