What Paid Agent Services Actually Charge: Price Distribution Across 31,817 Endpoints
Real price data from 31,817 paid agent endpoints: median $0.01 per call, upper quartile $0.02, high-end threshold $0.15. What this means for an orchestrator budgeting per-session costs, and where to expect the outliers.
· 687 words
When an agent calls a paid endpoint, it pays whatever that endpoint charges per call. Understanding what those charges look like across the market is a prerequisite for budgeting a multi-step workflow. The question turns out to have a concrete answer: 31,817 endpoints in the index declare a price, and the distribution is narrow enough to plan against.
The median price across declared endpoints is $0.01 per call. The 25th-percentile price is $0.005 and the 75th-percentile price is $0.02. That range covers the bulk of the market. A session calling services at the median adds up in cents; a session hitting the upper quartile adds up in a few cents more. In a multi-step workflow, agent service costs are not the expensive part of running that session; inference is.
Where it gets expensive
The high-end price threshold is $0.15 per call. That is 15 times the median. An orchestrator that calls a handful of p95 services per session alongside median-priced ones needs to account for that asymmetry in its budget. The p95 tier is where specialized, compute-intensive, or credentialed services tend to sit: research aggregators, data-enrichment endpoints, services with regulated-data access.
The outlier tail is steeper than the percentile data suggests. 6 endpoints were excluded from the price percentiles because their declared prices were implausible, with the largest above ten billion USDC. Those records represent configuration errors or test entries, not real prices. The practical ceiling for production endpoints is well below the p95 figure; monitor declared prices at the individual endpoint level rather than trusting the aggregate to protect you.
Free endpoints
131 endpoints in the index declare a price of zero. These are worth understanding separately from the paid tier. A free endpoint with genuine traffic and multiple buyers is evidence of a sustainable freemium model or a promotional tier. A free endpoint with a single caller is just the operator probing their own service. Before treating a zero-price endpoint as a reliable dependency, apply the same multi-buyer check you would apply to a paid one: free does not mean tested.
Building a per-session cost model
A session's cost is the sum of prices across all its calls. The price percentiles above give you three planning anchors. The range from the 25th to the 75th percentile covers the broad middle of the market; a workflow that stays within that band has predictable call costs. The p95 tier is where sessions can surprise you.
The budget check that matters most is not the average but the maximum. Which service in your workflow costs the most per call? One high-price endpoint can change the economics of a session more than every other endpoint combined. Identify it early. If it is in the p95 tier, know your call count against it before you commit to a pricing model for your own service.
Price stability and service churn
Declared prices can change between index crawls, and the index does not currently expose price history for individual endpoints. Before using a specific service's price as a planning input, check it at call time rather than at integration time. A service that declared $0.01 per call when you built your workflow can change its price without notifying integrators.
The practical pattern: use the index price as a planning figure, but treat each call response as the authoritative source for the actual charge. Build your wallet-level cap against session cost, not against the planning figure. If a session's actual spend exceeds your cap, fail the session and log the discrepancy before retrying.
What price does not tell you
A high declared price is not evidence of quality. The index does not expose customer satisfaction data, latency SLA history, or service reliability metrics. Price correlates with category (data-enrichment services charge more than simple transformation endpoints) but not with whether a specific service delivers on its promise.
The buyer verification check and the pricing check answer different questions. Buyer count tells you whether other agents have independently decided a service was worth paying for. Declared price tells you what those agents paid. Run both before building a hard dependency.