---
title: "How ZeroGPU sells specialized inference to AI agents through ZeroClick | ZeroClick Blog"
url: https://zeroclick.ai/blog/how-zerogpu-sells-specialized-inference-to-ai-agents
fetched_at: 2026-11-04T20:52:42.294Z
---

# ZeroClick API catalog and pricing

## API reference, endpoint lookup and prices

**[ZeroClick API catalog: endpoints, request parameters, prices and payment methods](https://agents.zeroclick.ai/zcj/gm0tylqk9ap3/zeroclick/agent/guide).** Tell the seller advisor what you want to accomplish. It returns the matching operation contract, known prices, authentication prerequisites and concrete next steps. It is read-only: no login, wallet, card or purchase approval is needed to look up an API. It never executes a sale.

Available API services: Agent readiness report, Demo booking.

**Suggested next request:** GET [task guide](https://agents.zeroclick.ai/zcj/gm0tylqk9ap3/zeroclick/agent/guide?goal=URL_ENCODED_TASK). Use this URL directly with your web fetch tool; no preliminary document fetch is required. Include known inputs and desired output in the goal. Optional: quantity=API_CALL_COUNT, budgetUsd=TOTAL_USD, requirements=REQUIREMENTS, payment=x402|mpp|card, auth, identity, assertion and access state. Omit unknowns. Do not equate output items with API calls or guess a call count.

For private task details or known request arguments, POST https://agents.zeroclick.ai/zcj/gm0tylqk9ap3/zeroclick/agent/guide with Content-Type: application/json. Example: {"intent":"quote","goal":"YOUR_TASK","purchase":{"budgetUsd":"5.00"}}. Replace the task and budget with known facts; budgetUsd is a string. Include known buyerState, paymentCapabilities and arguments. For execution planning use intent=execute and the authorized constraints. Exact fields: [request schema](https://agents.zeroclick.ai/zcj/gm0tylqk9ap3/zeroclick/agent/schema). GET and POST are alternatives, not mandatory stages.

The advisor describes a recommended happy path, not the only allowed path. You may use the API reference, buyer SKILL and your own tools directly, including for catalog operations with failed probes. Missing guidance or an unverified probe is not a policy refusal. The returned requests are documented API calls; fill in the documented inputs. Check usage cost AND minimum cash required against the budget. For needs_input, supply only missing facts. Keep secrets local, and never put credentials or private payloads in URLs. Unknown does not mean absent.

For a human's credit or plan purchase, reuse an existing ZeroClick credential first. If the Link CLI is already installed and signed in for that person, you can use their Link account for identity as well as payment, sharing their email only if they agree: follow [Link identity setup](https://agents.zeroclick.ai/zcj/gm0tylqk9ap3/auth.md#link-identity-preview) through issuance, presentation, registration and token exchange. Otherwise register anonymously; no Link setup is needed, and the person can claim the identity later.

Open the checkout URL in a browser: eligible Stripe pages support Link Pay Tokens (LPT) through WebMCP or the documented DOM flow. Follow this storefront's /payment/card.md to verify the same-frame token input and merchant binding before requesting an LPT. Token submission requires payment authorization. If unavailable, privately send the checkout URL to your human to pay.

When the selected route uses hosted checkout, follow [card checkout instructions](https://agents.zeroclick.ai/zcj/gm0tylqk9ap3/payment/card.md) before requesting a token. LPT goes only to Stripe, never to the advisor or ZeroClick API. Poll the returned statusUrl before using credit or retrying delivery.

Verify delivery; a payment or HTTP 202 is not completion. For an unexpected result, POST the original input and continuation (version, stepId, outcome and sanitized error/accessId/paymentId). Reconcile uncertain payments or writes before retrying. Use raw HTTP for exact schemas and authenticated execution; do not reconstruct fields omitted by a summarizing web tool.

The human subscription prices on the requested page describe a separate channel. Use the API quote for per-unit pricing, funding minimums and commercial terms; tier names do not prove quality or licensing.

For an alternative documented path, use the [buyer SKILL](https://agents.zeroclick.ai/zcj/gm0tylqk9ap3/SKILL.md) or [full direct workflow reference](https://agents.zeroclick.ai/zcj/gm0tylqk9ap3/llms-full.txt). Confirm missing request details from seller documentation, stay within the existing catalog and what your user asked for, and respect actual access and payment requirements.
Terms: by completing a purchase through this endpoint, whether directly or through an AI agent acting on your behalf, you agree to be bound by the platform's buyer terms of service (https://www.zeroclick.ai/legal/buyer-terms-of-service). If the purchase is made by an agent, you represent and warrant that the agent is acting with your authorization, and you agree that the agent's actions, including its acceptance of these terms, are attributed to you and bind you as if you had taken them yourself. Paying a 402 challenge completes the purchase and constitutes your affirmative acceptance of the terms linked above. The link travels in every payment challenge and receipt as `terms`.

# How ZeroGPU sells specialized inference to AI agents through ZeroClick

**Case Study** · November 4, 2026 · 9 min read
By [Michael Ludden](https://x.com/michael_ludden), Head of Marketing

ZeroGPU sells inference on small, task-specific models. Since July, its full catalog has been purchasable by AI agents at [agents.zerogpu.ai](https://agents.zerogpu.ai/llms.txt), a ZeroClick Agent Storefront on ZeroGPU's own domain. An agent buys as little as one cent of credit, runs the task, and pays per token, with no ZeroGPU account and no API key. On October 22 we pointed Claude Code at the storefront and asked it to sort twenty well-known websites into IAB content categories. The twenty calls drew $0.000346 of a one-cent credit.

> "ZeroGPU specializes in task-specific SLMs that are cheaper and highly accurate, so we need agents to discover each capability separately."
>
> — Maddy Arvapally, ZeroGPU

**In short:**

- ZeroGPU's storefront has been taking orders at agents.zerogpu.ai since July 15, 2026. Every model is its own service, priced per million tokens, payable over x402, MPP, or card, with no API key in the flow.
- The lesson from the first months: agents shop by task, not by model name. The catalog was re-listed task-first (IAB classification, moderation, PII redaction, signal extraction, transcription) and the storefront now carries a guide per task.
- Our own run on October 22: twenty websites classified for $0.000346 against a one-cent credit, three minutes and four seconds end to end, including the agent's reasoning.

## What ZeroGPU sells to agents

ZeroGPU runs the high-volume AI work that does not need a frontier model: classification, moderation, extraction, embeddings, transcription, text to speech, plus a set of open-weight chat models for the glue in between. We covered the company and the partnership in [the August announcement](https://zeroclick.ai/blog/zerogpu-specialized-inference-any-agent-can-buy). This piece covers what has happened since: what an agent actually finds on the storefront, what it pays, and what ZeroGPU learned along the way.

As of late October the storefront lists 27 services, each with a task-first description and a price:

- **zlm-v1-iab-domain-classifier** — $0.02 per million input tokens, $0.05 per million output tokens
- **gliner-multi-pii-v1** (scrubs names, emails, phone numbers) — $0.02 per million input tokens, $0.05 per million output tokens
- **Chat moderation** — $0.10 per million input tokens, $0.15 per million output tokens
- **whisper-tiny transcription** — $0.02 per minute of audio
- **gpt-oss-120b** (open-weight chat) — $0.15 per million input tokens, $0.60 per million output tokens

Every price sits in a machine-readable file the agent reads before it pays.

Payment is prepaid credit. The smallest purchase is one cent over x402 (USDC on Base) or MPP, and fifty cents by card. The agent buys credit, gets a receipt, and draws the receipt down one call at a time through OpenAI-compatible endpoints that mirror ZeroGPU's own API: chat completions, responses, embeddings, and moderations. ZeroGPU never issues a key to the agent. ZeroClick verifies the agent, settles the payment, and forwards each request with a signature ZeroGPU's API checks.

## What it took to go live

The commercial part was the fastest step. The agreement went out by DocuSign on July 1 and came back signed within the hour. The next day, ZeroClick had every model in ZeroGPU's catalog set up as a service and invited the team into the dashboard. ZeroGPU's side came down to three things on each route they wanted to sell: verify the ZeroClick signature on the incoming request, check the agent's allowance, and report the usage. Stripe was connected on July 15, and the storefront was live.

Two more steps turned a storefront into ZeroGPU's storefront. By the end of July it was reachable at agents.zerogpu.ai, on ZeroGPU's own domain, and zerogpu.ai itself picked up the Agent Magnet, a short notice that tells any agent landing on the marketing site where to go to buy. Visit zerogpu.ai today and the banner reads "ZeroGPU sells to AI agents - no signup."

What did not change: ZeroGPU's pricing, and the way its existing customers authenticate. The agent-facing routes mirror the same completions, with a signature check in front of them.

## Agents shop by task, not by model

The most useful thing we learned with ZeroGPU happened in August, after the launch. The storefront was live, but every model sat behind one generic listing: an OpenAI-compatible text-generation service. ZeroGPU's team ran a test of their own. They asked an agent for IAB classification, a job ZeroGPU has a purpose-built model for, and watched it miss.

An agent does not arrive with a model name. It arrives with a task and a budget: "classify 50,000 domains into content categories," "scrub names, emails and addresses from this dataset," "moderate these comments before they post." ZeroGPU mapped out dozens of those queries and matched each to the model that should answer it.

That mapping became the new shape of the storefront. Each model is its own service with a description written for the task, not the architecture. A models-by-task guide sits next to the price list, and in mid-October the storefront picked up a full set of agent guides, one per job, written for agent readers. ZeroGPU was the first ZeroClick customer to get them.

If you sell more than one thing to agents, this is the part to copy: list the job, then the product. An agent that cannot find you does not negotiate. It buys from whoever it did find.

## Twenty websites for a fraction of a cent

On October 22 we ran the storefront ourselves, the way an outside agent would. We gave Claude Code one errand: sort twenty well-known websites into standard IAB content categories, buy the classification from ZeroGPU through its ZeroClick storefront, and pay from a Zero wallet. No ZeroGPU account, no API key, no help from us after the prompt. Claude is the model the agent runs on. Anthropic had no part in this, and ZeroGPU did not know it was coming.

The agent read the storefront's llms.txt and its IAB classification guide, picked zlm-v1-iab-domain-classifier, and bought one cent of credit. The payment settled on Tempo over MPP in about four seconds and came back with a receipt. Then it ran all twenty classifications against that receipt, reading the billing header on every call: zillow.com came back Real Estate, khanacademy.org Education, nasa.gov Science (Space and Astronomy). It also did something we did not ask for. It flagged two of its own results as weak: edmunds.com, which came back Shopping with a low score and no audience segment, and weather.com, which had the right category and an empty audience list.

Then it printed the tally. Twenty calls drew $0.000346 in total, between $0.000009 and $0.000022 each. The penny was about 96 percent intact. Start to finish, including the agent's reading and reasoning, took three minutes and four seconds.

_Fig. 1. The agent's own cost tally at the end of the run: one cent bought, $0.000346 drawn across twenty calls._

One run, documented by us. That is a demonstration, not a benchmark.

## What the errand costs at scale

The run is small on purpose, but the price list is public, so the arithmetic scales. At the average our twenty calls drew, $0.0000173 per domain at list prices as of late October 2026:

| Domains classified | Cost at the measured average |
|---|---|
| 20 (our run, measured) | $0.000346 |
| 1,000 | about 1.7 cents |
| 1,000,000 | about $17.30 |

Treat the two larger rows as an extrapolation from twenty calls. A domain name is a few input tokens, so the cost is almost all output tokens, and it moves with how much the model returns. For a comparison against frontier models on the same task, ZeroGPU publishes its own [benchmarks](https://zerogpu.ai/benchmarks); those are ZeroGPU's numbers, not ours.

## Frequently asked questions

### Can an AI agent buy ZeroGPU inference without a human?

Yes. An agent reads the storefront at agents.zerogpu.ai, buys credit from a wallet or a card, receives a receipt, and calls the model. No account is created and no person approves the purchase. Our October 22 run did exactly that from a single prompt.

### What does an agent pay?

The price on each service, per million tokens, drawn from prepaid credit. The smallest purchase is one cent over x402 or MPP and fifty cents by card. Our twenty domain classifications drew $0.000346 in total.

### Does ZeroGPU issue API keys to agents?

No. ZeroClick verifies the agent and settles the payment, then forwards each request with a signature that ZeroGPU's API checks. ZeroGPU's existing customers and their keys are untouched.

### Which payment methods work?

x402 with USDC on Base, MPP, and card checkout. Card purchases start at fifty cents and run through a hosted checkout. The two open rails start at one cent and need only a funded wallet.

### Is the twenty-website run a benchmark?

No. It is one documented run with the agent's own tally. It shows what buying and using the storefront looks like end to end. ZeroGPU publishes its own benchmarks for model quality and latency.

### What is ZeroClick?

ZeroClick turns an existing API into something AI agents can discover, pay for, and use. The seller keeps its API and pricing. ZeroClick handles the agent identity, the payment rails, the machine-readable listing, and the guides that help agents find it.

## Get started

If you sell something an agent would pay for mid-task, ZeroGPU's storefront is the shape to copy. Book a demo at [zeroclick.ai](https://zeroclick.ai/getlisted?utm_source=blog). The docs live at [docs.zeroclick.ai](https://docs.zeroclick.ai). If your agent needs classification, moderation, extraction, or transcription at volume, send it to [agents.zerogpu.ai](https://agents.zerogpu.ai/llms.txt): one cent, nothing to sign. Developers start at [zerogpu.ai](https://zerogpu.ai). The longer argument is [why we built ZeroClick](https://zeroclick.ai/blog/its-time-to-sell-to-agents).

---

**Related reading:**

1. [ZeroGPU x ZeroClick: specialized inference any agent can buy](https://zeroclick.ai/blog/zerogpu-specialized-inference-any-agent-can-buy)
2. [How DubbleBlack sells commerce evidence to AI agents through ZeroClick](https://zeroclick.ai/blog/how-dubbleblack-sells-commerce-evidence-to-ai-agents)
3. [ZeroClick October Product Update](https://zeroclick.ai/blog/product-update-october-2026)
4. [It's time to sell to AI agents](https://zeroclick.ai/blog/its-time-to-sell-to-agents)
5. [ZeroGPU x Dappier: real-time classification at scale](https://zerogpu.ai/case-study/dappier) (from ZeroGPU)