Inference pricing
What an inference costs in KVR today, why a decentralized network is structurally cheaper, and how price is meant to fall as supply grows.
Access to the network is pay-per-inference: the gateway quotes a KVR price for your request, your wallet pays it on-chain, and only then does the hub run the model. Pricing is a small, transparent formula — a per-request floor plus a per-token rate — quoted up front and settled on actual token usage after generation.
The settlement formula — quoted before, charged on real usage after.
cost (KVR) = basePrice + total_tokens × perToken # quote: estimate with the model's nominal output length # charge: recompute on the real prompt + completion tokens
No central margin to pay for
A centralized API prices at cost plus a large margin and capital recovery. A decentralized network prices near its contributors' marginal cost — electricity and hardware amortization — plus a thin protocol fee. That structural gap exists regardless of size. Growth widens it: expert-sharding means more nodes each hold a smaller slice, so cheaper devices can serve, lowering the marginal cost of participation and deepening supply.
Price is governed, not a free-for-all. Rates are a sensitive economic parameter, changed only by the genesis wallet under wallet-signature + 2FA — never by an environment variable. This keeps the token economy stable and auditable.
Utilization-driven price
The design direction is a price that floats with network utilization between a floor (kept above node marginal cost, so serving stays worthwhile) and a ceiling (kept below centralized alternatives, so it stays competitive). Idle supply nudges the price down; congestion nudges it up. That is the mechanism that finally makes "more shared nodes → lower price" true in code — the natural thermostat of the KVR economy.