Arbius is a decentralized machine learning network and token that generates new tokens by participating in GPU computation within the network. Arbius lacks a central governing body for creating new tokens, ensuring complete openness. Holders can vote on-chain for protocol upgrades, and the model operates as a DAO, providing income for model creators. Arbius is user-controlled and avoids monopolization by large corporations and governments. It is designed to make it difficult or impossible to censor its use, allowing anyone worldwide to interact with AI models without permission. The total supply of Arbius tokens is 1 million, minted through task rewards, with a continuous halving process. Task rewards are adjusted based on the ratio of the total supply to the expected total supply. Rewards decrease according to the utilization of the overall GPU computational capacity. 10% of Arbius will be sent to the Arbius DAO, while another 10% will be provided to those who stake DML/ETH LP tokens. Arbius is currently in its early experimental phase and does not imply future profitability expectations.