67.3% of x402 Endpoints Require Caution Before Payment

Lumière PayCheck has identified that the majority of accessible AI API endpoints currently fall into a "caution" category, meaning their delivery reliability has not yet been confirmed by paid testing. This data highlights a critical gap for developers building autonomous agents: while many APIs accept payment quotes, they do not guarantee successful execution until a transaction is completed.

The Landscape of Verified Reliability

The latest monitoring cycle provides a snapshot of the current state of machine-to-machine commerce on the x402 protocol. As of November 1, 2026, PayCheck is actively watching 108,624 endpoints listed in the x402 Bazaar. The distribution of verdicts reveals that only a small fraction of these endpoints have been fully verified for reliable delivery after payment.

Of the total endpoints monitored, 21,443 (19.7%) received a "proceed" verdict. This classification indicates that the endpoint responds reliably with a valid payment quote. However, it is important to distinguish this from confirmed delivery. A "proceed" status means the API is technically ready to accept payment, but it does not guarantee that the service will be delivered successfully after the transaction.

In contrast, 73,110 endpoints (67.3%) are marked as "caution." This verdict primarily signifies that delivery has not been confirmed by a paid test. Many of these endpoints may require higher payment amounts than our tester can provide, or they might necessitate user sign-in processes that automated agents cannot complete. This label is not an indication that the endpoint is broken or unsafe. Rather, it signals that the agent builder should exercise additional verification before committing funds.

A smaller subset, 3,160 endpoints (2.9%), received an "avoid" verdict. These endpoints failed specific checks, such as being down, returning broken payment quotes, or failing to deliver content during a paid test. Additionally, 10,851 endpoints had not enough data for a reliable verdict, and 60 were identified as free services that do not require payment.

For developers, the takeaway is clear: the majority of available APIs are technically functional but lack confirmed delivery proof. Relying solely on the presence of a valid quote is insufficient for building robust agent workflows. Builders should prioritize endpoints with higher confidence levels and implement strict spending limits to mitigate the risk of undelivered services.

Confidence in the Top Tier

While the broader landscape shows significant uncertainty, the highest-scoring endpoints demonstrate that reliable machine-to-machine payment is achievable. Of the top 100 endpoints by score, 100 have delivery verified by a real paid test. This perfect verification rate among the best performers suggests that when an endpoint is properly configured and priced correctly for automated agents, it can deliver consistent results.

This distinction underscores the importance of curation. The x402 Bazaar lists over 100,000 endpoints, but only a small fraction have been rigorously tested through actual financial transactions. The gap between the 19.7% "proceed" verdicts and the fully verified top tier highlights that valid payment quotes are common, but guaranteed delivery is rare.

For agent builders, this means that finding reliable partners requires looking beyond the initial quote. Endpoints in the top tier have proven they can handle the full cycle of payment and service delivery. Integrating with these verified endpoints reduces the operational overhead of handling failed transactions or debugging silent failures where an API accepts money but provides no output.

How We Measure Reliability

Lumière PayCheck operates by continuously monitoring every endpoint listed in the x402 Bazaar. The system checks these endpoints on a regular schedule to assess their availability and responsiveness. A key component of this process involves making small, real paid test calls to verify that the service is delivered as promised after payment.

The infrastructure for this verification relies on reading real settlements on the Base blockchain. This ensures that the data reflects actual financial transactions rather than simulated or theoretical scenarios. The latest monitoring cycle processed 22,321 checks in just 35 minutes, with 21,762 answered correctly (97.5%). This high success rate in processing checks demonstrates the system's ability to keep up with the dynamic nature of the x402 ecosystem.

By combining quote validation with actual payment testing, PayCheck provides a more accurate picture of endpoint reliability than quote analysis alone. This method allows developers to distinguish between APIs that are merely configured for payments and those that are proven to deliver value.

What to Do

To build resilient AI agents that interact safely with the x402 economy, developers should adopt the following practices:

* Check before paying: Always query Lumière PayCheck for the current verdict of an endpoint before initiating a payment. Relying on static configurations can lead to failures if an endpoint's status changes or if it lacks sufficient verification data. * Set spend limits: Configure your agents with strict budget caps for individual API calls. This protects your application from unexpected costs, especially when interacting with endpoints that may have variable pricing or require sign-in processes. * Prefer verified endpoints: Prioritize integrating with endpoints that have been confirmed through paid tests. These endpoints offer the highest assurance of delivery and reduce the need for complex error handling related to undelivered services.

The full live list is at https://lumierepaycheck.org/stats and the API is free at https://lumierepaycheck.org/v1/stats