One million output tokens from a frontier model. The unit of intelligence.
- Peg pool
- —
- Feed
- —not deployed · 6h window
- Coverage
- —
- Markets
- 0
Four units a token can be paired with. Three are synthetic units of model output the protocol mints and redeems at the index; one is Venice AI's compute token.
Synthetic. The protocol mints each one into a single-sided ask at the index price and holds the dollars buyers paid in the vault. All three redeem for USDC or DIEM, at the index, in the same transaction.
One million output tokens from a frontier model. The unit of intelligence.
One billion output tokens from an open-weight model. Intelligence at cost.
One hour of an agent working: 1.6M tokens read, 400k written, at frontier prices.
The price of intelligence, written on-chain every hour by a Chainlink Functions network from OpenRouter's public prices. A seven-day median of hourly samples. Nobody edits it.
Every inference unit in circulation is backed by dollars and DIEM the vault holds, plus the peg's bid. Coverage is that backing divided by what the outstanding unit is worth at the index, computed on-chain from live feeds so anyone can check it. It rises as inference gets cheaper, and two permissionless calls keep it where it should be.
DIEM is Venice AI's own ERC-20 on Base, routed through Aerodrome, and a pair coin in its own right. It is also what the vault delivers when you redeem an inference unit for inference.