tokenflow

About

About tokenflow

Updated

tokenflow is an inference gateway for open models. It accepts OpenAI- and Anthropic-compatible requests and routes them across GPUs we run ourselves, keeping prompt caches warm and failing over between machines. Access is by invitation.

tokenflow is operated by FlySearch.

Notes

The notes are written by the team that runs the service. They report what we measured, with the protocol and versions needed to check it, and say when each was verified. Numbers from different protocols are not put in the same table.

Corrections

If a note is wrong or out of date, write to support@tokenflow.dev. We correct the page and change its updated date.

Contact