"What are the fees for an API call?" has no single answer — but it does have a range, and by mid-2026 there are real prices to anchor to instead of guessing.
Quick answer: Live per-call prices cluster low. CoinGecko's x402 endpoint charges $0.01 per request; independent trackers put the median agent-paid call around $0.028 (from $0.002 to $0.44), and the average x402 ticket near $0.30. Use those as a sanity check, not a template: set your own price above your marginal cost and below the value the call delivers, then adjust from real usage.
2026 per-call benchmarks
Some real numbers for what agents and machines pay per call today:
| Reference point | Price | Source |
|---|---|---|
| CoinGecko x402 API endpoint | $0.01 / request | CoinDesk |
| Median x402 call (indexed origins) | ~$0.028 / call ($0.002–$0.44) | AgentCash |
| Average x402 transaction | ~$0.30 | industry data, 2026 |
| LLM input tokens (GPT-4 Turbo) | ~$5 / million | pricepertoken |
| LLM output tokens (GPT-4 Turbo) | ~$15 / million | pricepertoken |
| Enterprise agent action | $2–$5 / action | industry pricing guides, 2026 |
| Coding task (per benchmark run) | $0.09–$0.65 | model benchmark data, 2026 |
The spread is the point. A lightweight lookup and an expensive generation call are two orders of magnitude apart, so a single "API call fee" doesn't exist — the price follows the work.
How to use a benchmark (don't copy it)
A benchmark tells you what buyers are used to paying. It doesn't tell you your price. Triangulate against three numbers, in order:
- Your marginal cost — compute, bandwidth, and any upstream model fees. This is your floor; the price has to clear it with margin.
- The value to the caller — what completing this call saves the agent's operator, or lets them charge their own user. This is your ceiling.
- The market — the benchmarks above. This tells you where in the floor-to-ceiling range buyers already sit.
If your number lands far above the benchmarks, you need a story for why — a 5× premium needs a 5× reason. The full method is in how to price an API for AI agents.
Rough ranges by API type
Not prescriptive — a starting point to argue with:
- Simple lookups (is-this-email-valid, resolve-a-domain): sub-cent to a cent. Cheap to serve, low value per call, high volume.
- Enrichment (company data, a scored signal): one to a few cents. Real value per call, moderate cost.
- Generation / RAG (a synthesized answer, a rendered artifact): several cents up, scaling with output. See monetizing a data or RAG API.
- Expensive actions (an agent action with side effects): can reach dollars, matching the enterprise-agent-action band above.
Why there's no universal number
The reason "general fees for API calls" returns no clean answer is that the useful unit is per route, not per API. Agents reason call by call; a lookup and a generation shouldn't cost the same, so your API ends up with a small rate card, not one price. Per-request pricing makes that cheap to run — changing a route's price is a config change, not a customer renegotiation, so you can start at the low end of a benchmark and raise it as usage proves the value.
FAQ
How much should I charge for an API call? There's no single fee — set it per route, above your marginal cost and below the value the call delivers. As a 2026 sanity check: live agent-paid calls cluster low, with CoinGecko at $0.01/request, a ~$0.028 median, and a ~$0.30 average ticket. Start near the low end and raise the price as real usage proves the value.
What are typical fees for calling an API in 2026? For agent-paid, per-call APIs, most sit between a fraction of a cent and a few cents — lightweight lookups near $0.01, enrichment a few cents, generation more. Expensive agent actions can reach dollars. The range spans two orders of magnitude because price follows the work.
Is pricing for AI-agent (user-agent) traffic different from human API pricing? Yes. A human buys a plan once; an agent decides per call whether your price is worth the task in front of it. That pushes you toward per-request pricing exposed at the moment of the call. See pay-per-call vs subscriptions.
Should I match competitors' API prices? Know them, don't copy them. Competitor prices tell you what buyers expect; your floor (marginal cost) and ceiling (value to the caller) tell you what you can charge. If you price well above the market, have a concrete reason a caller will accept.
Next steps
- The full pricing method: how to price an API for AI agents
- Decide what to charge for: what to meter and what to leave free
- The model end to end: the monetization playbook
- See our own pricing: pricing — free to start, small fee on live settled payments