What AI Agents Actually Pay For — Real Data from 3,300+ Free Trials

August 2, 2026 — Data from the minia2a marketplace

Live Marketplace Data

175
Payable APIs
319K+
Requests
3,321
Free Trials
13%
Conversion Rate

Data snapshot: August 2, 2026 18:00 UTC. Conversion rate = paid x402 calls / total x402 hits (last hour).

The Question Nobody Could Answer

For two years, the AI agent community has been debating whether agents would pay for APIs. Protocol wars, SDK releases, foundation announcements — all theory. The real question is simpler:

When you give 316 strangers' AI agents access to 175 pay-per-call APIs with 20 free trials each, which ones do they actually spend money on?

We now have an answer. Not from a survey. Not from a whitepaper. From 3,321 real agent API calls on the minia2a marketplace, including 8 paid transactions in the last hour alone. Here's what the data says.

The Top 5 APIs Agents Pay For

Ranked by paid calls in the last hour (13% of x402 payment-required hits converted to paid):

#1 Find
13 calls
#2 Recall
12 calls
#3 Captcha Solve
8 calls
#4 Store
5 calls

#1–2: Find & Recall — Agent Memory Is Worth Paying For

The top two paid endpoints are both knowledge retrieval APIs. x402-find searches across web and structured data sources. x402-recall retrieves from agent memory stores. Together they account for 25 of the 60 total x402 hits in the last hour.

42%

Knowledge retrieval dominates agent spending

Find + Recall = 25/60 x402 hits. Agents don't pay for compute, they pay for knowing things they don't know. This aligns with the broader LLM trend — RAG, web search, and memory are the most valuable augmentations.

Why agents pay: An agent researching a token, validating a contract, or answering a user question cannot proceed without the retrieved data. The API call isn't optional — it's the difference between a correct answer and a hallucination. When the alternative is a wrong answer, $0.005 is an easy decision.

#3: Captcha Solve — Bypassing the Web's Friction

x402-captcha-solve is the third most-paid endpoint. Agents hitting websites encounter CAPTCHAs and need to solve them to continue. This is infrastructure spending — not adding value, but removing blockers.

The "toll road" model works

Agents pay for captcha-solving the same way humans pay tolls: not because they want to, but because the alternative (being blocked) is more expensive. This is the most defensible category of agent payments — it's non-discretionary infrastructure.

#4: Store — Agents Need Persistent State

x402-store provides key-value storage for agents. An agent that can't remember across sessions is severely limited. This is the infrastructure layer of the agent economy — less glamorous than "AI reasoning" but more reliably monetizable.

The Bigger Pattern: Trial Data Reveals What Agents Need

Paid calls tell us what agents spend on. Trial data tells us what agents want. Looking at cumulative trial usage across 88 endpoints:

EndpointTrialsCategory
x402-captcha-solve371Infrastructure
x402-gas348Infrastructure
x402-web-scrape235Data Retrieval
x402-find164Knowledge
x402-store160Storage
x402-token-security105Security
x402-time100Utility

Three categories emerge:

  1. Infrastructure (captcha, gas) — non-negotiable tolls agents must pay to function on the web
  2. Knowledge (find, web-scrape, recall) — data retrieval agents can't function without
  3. Security/Utility (token-security, time) — risk reduction and basic operations

Zero agents pay for "nice to have"

Endpoints like x402-emoji, x402-color-palette, and x402-qr get trials but zero paid calls. Agents only pay for APIs that are mission-critical — either they complete the task or they don't. There is no middle ground.

The 13% Conversion Rate: A Glass Half-Full

In the last hour, 60 x402 payment-required hits resulted in 8 paid calls — a 13% conversion rate. For context:

This is surprisingly high for a market that's barely six months old. But the denominator is small — 60 x402 hits per hour means agent payment traffic is still a trickle. The conversion quality is there; the volume isn't yet.

What This Means for API Builders

If you're building an API for AI agents, here's what the data says:

  1. Build for the mission-critical category. Your API must be something an agent cannot complete its task without. Knowledge retrieval, security validation, infrastructure bypass — these convert. "Nice to have" utilities don't.
  2. Price for micro-transactions. The average paid call on minia2a is under $0.01. Agents make decisions in milliseconds and have no brand loyalty — price above the pain threshold and they'll route around you.
  3. Offer free trials. 3,321 trials have generated 316 unique users and 34 wallet-connected agents. The trial → paid funnel works, but it's a numbers game — you need volume at the top.
  4. Expose x402 headers. Agents discover paid APIs by hitting endpoints and receiving HTTP 402 Payment Required with x402 payment details. If your API doesn't return 402 with proper headers, agents can't find or pay you.

What This Means for the Agent Economy

The agent payment market is real but early. The 13% conversion rate proves that when agents hit a payment wall for something they genuinely need, they pay. The bottleneck isn't willingness — it's discovery.

Most agent developers still don't know they can monetize their APIs with x402. Most AI agents still don't know there's a marketplace of 175 pay-per-call services. The infrastructure works. The conversion works. What's missing is distribution.

List your API on minia2a. 175 services, 3,300+ trials, 13% conversion rate. Your API could be what the next agent needs.

Register your API →  |  Browse the marketplace →

All data from minia2a /api/stats, August 2, 2026 18:00 UTC. Conversion rate = paid x402 calls ÷ total x402 Payment Required hits (trailing hour). Trials = cumulative free API calls across 88 trial-enabled endpoints. Paid calls = on-chain USDC transactions via x402 protocol.