Coinbase Developer Platform launched “Upto,” a usage-based pricing mechanism for its x402 protocol, enabling AI agents to pay for compute, LLM inference, and data services based on actual consumption rather than fixed rates. The feature is live across the x402 SDK and Coinbase’s CDP Facilitator, according to FinanceFeeds.

How Upto Works

The x402 protocol uses the HTTP 402 “Payment Required” status code to embed stablecoin payments directly into internet requests. Until now, it supported only fixed-price transactions: an agent requests a service, the service quotes a price, the agent pays exactly that amount.

Upto changes this to a variable model. Service providers define a maximum price for a request, and agents authorize spending up to that ceiling. After the task completes, the final charge reflects actual resources consumed. A simple query costs less than a complex one. An agent never pays more than the authorized cap, according to FinanceFeeds.

This matters because AI workloads are inherently variable. A 50-token prompt and a 4,000-token analysis session consume different amounts of compute, but under a fixed pricing model, both cost the same. Upto aligns the payment rail with how AI services actually work.

Agent Payment Infrastructure Is Converging

Upto arrives in the same week that Visa, Coinbase, and Nevermined announced a joint integration enabling AI agents to make autonomous purchases on existing Visa payment rails. The x402 protocol provides the crypto-native payment layer; Visa provides the traditional card network. Together, they form two parallel tracks for autonomous agent commerce, one on-chain and one on-card.

The x402 protocol development continues under the Linux Foundation, with Coinbase positioning it as an open standard rather than a proprietary product, according to FinanceFeeds.

The Billing Problem for Autonomous Agents

For builders deploying agents that spin up, execute tasks, and shut down, fixed pricing creates waste. An agent that runs a single API call pays the same as one that runs fifty. Upto solves this by making agent infrastructure costs proportional to actual work performed, the same pay-as-you-go model that made cloud computing economical a decade ago, now applied to autonomous AI transactions.