Retired inference engines

FreeToken removal

Removed on 9 October 2026.

The experimental FreeToken engine has been removed. It is no longer offered in Models or Services. Its runtime image build/promotion workflows and active regression cases have also been removed. Use Ollama or vLLM for new local models.

Previously saved FreeToken definitions remain visible. They cannot be edited, started, or restarted; the API rejects these operations and the controller does not create replacement workloads. Stop and Remove still clean up their old managed runtime. Until stopped, their saved reservations and existing Pods continue to protect GPU capacity and shared model caches. Stopping the old Pod also releases its temporary download cache.

Choose a supported engine and create a new definition to continue using a model. Changing an engine does not guarantee the same artifact or quantization is supported. Check the chosen runtime's model requirements.

Dated FreeToken reports and screenshots describe earlier versions only.