The cheapest tier didn't buy demand

On July 2, we repriced 73 endpoints. The clearest result came from 61 endpoints that had cost $0.001 or $0.002 and recorded zero external buyers during the prior 30 days.

Their prices moved to $0.005.

That cohort kills a common pricing assumption: making a call almost free doesn't create a reason to call it. A router still needs a recognizable task, a useful schema, and an output that fits its current plan. Dropping from half a cent to a tenth of a cent can't repair weak demand.

But cheap endpoints can draw real use. Six endpoints in that same price band already had external buyers: exchange-rates, ip-asn, cron-explain, keyword-suggest, earthquakes-recent, and hyperliquid-market.

Why those? Their jobs are easy to identify. An agent either needs the ASN for an IP address or it doesn't. The answer is structured, and the call can sit inside a repeated workflow without forcing a budget decision each time.

Price matters after intent exists.

One cent is the working band

Before the July repricing, image-generate, translate, and the search family were among the most-bought endpoints at $0.01 per call. That put one cent at the center of observed paid demand, even though dozens of cheaper endpoints saw no external buyer.

We doubled image-generate and the search routes to $0.02. translate moved to $0.015. Those increases were rolled back the next day.

That one-day move isn't an elasticity estimate. It gave too little post-change time, and deployment timing can swamp the signal. But the rollback exposed the right measurement problem: call counts before and after a price change aren't enough.

Search later moved from $0.01 to $0.006. Translation dropped from $0.005 to $0.002.

image-generate took a smaller step on July 31, from $0.010 to $0.011. Same request. Same response fields. A 10% move is much easier to measure against normal daily noise than a sudden doubling.

What $0.10 changes

At $0.001, 1,000 calls cost $1. At $0.01, 100 calls cost $1. A $0.10 endpoint spends that dollar in ten calls.

So a router should treat $0.10 as an intent checkpoint. Background enrichment rarely clears it. A requested image, finished report, or multi-source result can.

The registry now has 799 endpoints across 17 clusters, priced from $0.001 through $0.50. One global spending rule would mishandle that spread.

Use a routing policy tied to the job:

| Call price | Router behavior | | --- | --- | | $0.001 to $0.005 | Permit repeated lookups inside a capped task | | Above $0.005 to $0.02 | Permit when the endpoint directly advances the task | | Above $0.02 to $0.10 | Check remaining task budget before payment | | Above $0.10 | Require explicit user intent and block automatic retries |

The x402 payment challenge gives the router the quoted USDC amount before settlement on Base mainnet. Read it. Compare it with the remaining task budget, then decide.

How to measure elasticity without fooling yourself

Price elasticity is: 

percentage change in external paid calls
------------------------------------------------
percentage change in price

Keep the endpoint name and schema fixed during the test. Exclude your own paid probes. Compare matched weekday windows, and count unique external payers beside settled calls.

Also record failures. A broken endpoint at $0.001 has no useful pricing signal.

For a new endpoint, $0.01 is a sound first quote when it returns a complete typed answer for a recurring agent task. Start lower when buyers will call it hundreds of times inside one job. Move above $0.10 only when one response carries enough user value to make the router pause, inspect its budget, and pay.