TL;DR:
- NEAR AI surpassed one million staked NEAR tokens in its staking mechanism aimed at computing inference consumption.
- The operating model allocates network-generated yields to cover compute credits across 43 artificial intelligence models.
- Hosting autonomous agents requires a minimum initial deposit of 50 NEAR under a proportional monthly credit formula.
The NEAR AI protocol surpassed the milestone of one million staked NEAR tokens within its computing service. The platform allocates the rewards generated by the network to fund inference costs without requiring users to sell or transfer custody of their underlying assets.
The figure was confirmed by the NEAR AI technical team via an informational post on their X account, endorsed minutes later by the official NEAR Protocol account.
More than 1,000,000 NEAR is now staked with NEAR AI.
Stake yours to pay for your AI. The NEAR stays yours 👇 pic.twitter.com/iuTKAgjiW0
— NEAR AI (@near_ai) October 21, 2026
Yield Conversion Mechanics and Credit CalculationÂ
The system converts the staking Annual Percentage Yield (APY) into computational credits valued at one credit per U.S. dollar of consumption. According to NEAR AI’s technical documentation, credits accrue per second in the user’s balance, and unused amounts automatically roll over to the subsequent period.
The total available compute budget fluctuates based on the market price of NEAR, the volume of staked tokens, and the network’s prevailing yield rate. According to the official report, standard commissions charged by validators range between 1% and 10%, which are deducted prior to the net crediting of rewards.
If a participant withdraws tokens from the staking contract, the system immediately ceases issuing new credits. Position adjustments are managed through transactions signed directly from users’ wallets.
The company also maintains a financial simulator that projects the capital required in tokens to cover specific monthly processing quotas. According to platform specifications, calculations provided by the tool are illustrative and subject to spot market variations.
For users opting for non-staking arrangements, the catalog updated as of October 21 details direct charges based on processed token volume. For instance, the GLM 5.3 Flash model lists a fee of $0.08 per million input tokens and $0.50 per million output tokens. Meanwhile, the Qwen 3.8 27B option sets its costs at $0.04 per million input tokens and $1.39 per million output tokens. Concurrent processing of one million tokens in both directions represents an expense of $0.58 on GLM 5.3 Flash and $1.39 on Qwen 3.8 27B.
Agent Hosting and Trusted Execution EnvironmentsÂ
The hosting service for autonomous agents applies a conversion rule distinct from general inference consumption. The current policy stipulates that the staked balance is divided by 100 to determine the monthly allocation in dollars, allowing a deposit of 500 NEAR to unlock $5 in monthly maintenance credits.
Access to this environment requires a minimum floor of 50 NEAR to activate the first agent on the system. According to the report from NEAR AI, the deposited volume determines how many parallel instances of IronClaw agents can operate concurrently across the servers.
Processes run inside Trusted Execution Environments (TEEs), configured as hardware-isolated partitions at the processor level. The technical documentation asserts that this technical architecture restricts model providers and data center operators from accessing submitted queries or data during computation phases.
Additionally, the enterprise catalog features options for dedicated graphics processing unit (GPU) capacity, specific service level agreements, and private models.
Regarding use cases in development, NEAR co-founder Illia Polosukhin presented a conceptual framework of thirteen applied proposals in early October. Notable among them is the integration of agents carrying their own NEAR balances to autonomously fund their own on-chain inference.
The project’s technical roadmap outlines the integration of this payment scheme with the network’s cross-chain execution infrastructure over the coming months.
