NEAR AI Hits Milestone as Users Stake Over 1 Million Tokens for Inference

The mechanism, introduced on July 30, enables users to convert staking rewards into credits for AI inference. Because the staked NEAR remains in the user’s wallet, the platform creates a self-sustaining loop where the network's yield covers the costs of interacting with models from providers like OpenAI, Anthropic, and Google. Users accumulate credits second-by-second, with unused balances carrying forward as long as the tokens remain staked.

Budgeting for these services requires balancing token price volatility with staking yields. While inference credits accumulate via yield, agent hosting operates on a distinct policy: every 500 NEAR staked currently generates $5 in monthly credits. This system also dictates the capacity for running IronClaw agents, requiring a minimum of 50 NEAR to initiate operations. Beyond these automated staking options, the platform maintains a pay-as-you-go model for direct token-usage billing, providing an alternative for those who prefer not to lock their assets.

NEAR co-founder Illia Polosukhin has recently proposed expanding these capabilities to include autonomous agents, such as NFT-based digital pets that manage their own NEAR balances to pay for tasks. By integrating trusted execution environments, the protocol aims to offer private inference, ensuring that prompts and data remain shielded from cloud operators and model providers during processing.