Polygon Labs just dropped a striking benchmark: "Agent Pay Channels," an x402-based system for AI agents paying for inference, data, and API calls as they consume them. The headline: 11M+ verified payment updates per second across 25 hubs.
Here's the honest breakdown.
What 11M/sec actually means
The 11M figure counts offchain channel updates — not onchain transactions. Individual payment updates move through payment channels offchain; deposited funds stay secured on Polygon and accumulated states settle onchain in batches. The hubs partition payers and don't coordinate with each other, so capacity scales by adding hubs (Polygon estimates 100M+/sec with a larger fleet).
The numbers that matter
Full x402 path (agent + site + facilitator + hub): ≈ 40,000 payments/sec — 2.4M payments recorded, 100% success rate
Single hub, engine-direct: 533,000–536,000 fully verified payments/sec on one 24-core server
Confirmation: ~20 microseconds per update (excluding network latency)
Cost: ~$0.15 per billion payment updates
Test setup: OpenRouter-style inference API on a live devnet, signed payment triggered every 100-token window (payment path real, inference provider a stand-in)
Settlement: configurable — after 1 payment, 50k updates, or 100M updates
Why it matters for x402 builders
When machine payments get this cheap and this fast, the bottleneck moves. The question stops being "can the agent pay" and becomes "should this agent pay for this" — a decision model's confidence score gating whether the x402 payment fires at all.
That's the pattern I'm building around: decision-gated machine payments. The gate is the product, not the model.
Sources: crypto.news · Traders Union
Top comments (0)