The GPU network
Where the compute comes from, and what Eterial's relationship to it is.
The models are served by Gonka, a decentralised network in which independent operators contribute GPU capacity and are paid in the network's own token, GNK. Because capacity comes from operators competing to supply it rather than from one company pricing it, inference on the network costs a fraction of what a conventional provider charges.
The network is Gonka's, not ours, and it is documented by them — start at gonka.ai if you want the protocol itself.
The two things it changes for you
Prices move. Rates on the network are set by its governance and are polled periodically, so the cost of a model is not a number fixed in a contract. See Pricing.
A cost can be corrected after the fact. Payment is in GNK while your balance is in dollars, so a request is charged at the exchange rate captured when it ran and reconciled once the network confirms the actual GNK amount. Where those differ, a correcting entry appears on your balance — which is what a provisional cost in the usage log becomes. See Billing.
When the network cannot serve a request
It falls back to OpenRouter, automatically, inside the same request. See Failover & retries.