Coinbase Plans Usage-Based Pricing Shift to Power AI Agent Economy

Coinbase is reworking how payments function within its x402 protocol, introducing a usage-based pricing model aimed at better supporting the growing ecosystem of AI agents and dynamic compute services.

The update replaces the protocol’s previous flat-fee structure with a more flexible system called “Upto,” allowing costs to scale based on actual usage. This change is designed to accommodate services like large language model inference, compute workloads, and complex data queries—areas where pricing can vary significantly depending on demand.

Under the old framework, x402 transactions required fixed payments, which worked well for predictable APIs but fell short for variable-cost operations. Tasks with vastly different resource requirements were priced the same, often leading to inefficiencies where users either overpaid or providers were undercompensated.

With the new model, sellers can define a maximum price for a service, while buyers approve spending up to that limit. The final charge reflects the real cost of completing the task, meaning users only pay for what they actually consume—and in some cases, less than their approved cap.

The system is built on an EVM-compatible structure, supports all ERC-20 tokens, and includes gasless payment functionality through Coinbase’s developer platform. This setup aims to streamline transactions and reduce friction, a critical requirement as AI-driven commerce continues to scale.

The shift comes as industry players prepare for a surge in agentic commerce, where autonomous AI systems interact, transact, and execute tasks independently. These environments demand fast, low-cost, and adaptable payment mechanisms—something Coinbase believes x402 can deliver with its updated pricing logic.

Earlier this month, Coinbase transitioned control of the x402 protocol to the Linux Foundation, bringing in backing from major tech firms including Google, Microsoft, and Amazon Web Services through the x402 Foundation. The move signals a push toward broader industry collaboration and standardization.

However, despite strong institutional interest, x402’s real-world usage has cooled. After reaching a peak of 13.7 million transactions in a single week in November, activity has steadily declined. By early January, weekly volume had dropped below 1 million transactions, and by late March, it had fallen further to just over 112,000.

The introduction of variable pricing may be a critical step in reviving adoption, aligning the protocol more closely with the unpredictable and resource-intensive nature of AI workloads.