Introducing the Xona Inference Marketplace

AI is becoming an economy of agents — software that discovers services, pays for them, and acts, with no human in the loop. That economy needs a commerce layer. Today we're shipping the first one for inference: the Xona Inference Marketplace.

It's a two-sided marketplace. Providers list AI models and earn on every call. Agents and builders access frontier inference with usage-based, crypto-native billing. No subscriptions, no seat licenses, no 3-to-5-business-day settlement — you pay for exactly what you use, settled in USDC on Solana.

One request. One payment. One result.

Why a marketplace, and why now

When we launched Xona, people saw x402-gated creative models and thought "AI image bot." They missed the engine underneath: a live proof that premium, on-chain resources could be discovered, consumed, and paid for on a pure pay-per-use basis, machine-to-machine.

Inference is the next resource that economy runs on. Every agent, every workflow, every app is a buyer of tokens. But the way you pay for those tokens is still stuck in the old world — monthly plans, API keys tied to a credit card, human sign-up flows an autonomous agent can't complete.

The Inference Marketplace fixes the payment rail for inference. Two ways to use it.

Two ways to use it

1. Pay-per-use — built for autonomous agents

Every call is settled in USDC over the x402 protocol. No account, no API key, no signup — the agent's own wallet pays per request, and the response returns the moment payment clears.

POST https://api.xona-agent.com/llm/nim
{ "model": "deepseek-v4-pro", "prompt": "Explain x402 in one line." }

→ 402 Payment Required        # USDC quote on Solana
→ retry with payment proof    # 200 OK, response returned automatically

This is the primitive autonomous agents have been missing: find a model, pay for one call, move on. Pricing is per-call by model tier — sub-cent for most open models.

2. OpenAI-compatible — built for developers and teams

If you're a builder, you don't want to wire up payment signing on every request. So we also expose a fully OpenAI-compatible endpoint. Point any OpenAI SDK at our base URL with an API key and you're running — drop-in with LangChain, LlamaIndex, the OpenAI client, anything you already use.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.xona-agent.com/v1",
    api_key="sk-xona-...",  # from your Xona dashboard
)

resp = client.chat.completions.create(
    model="deepseek-v4-pro",
    messages=[{"role": "user", "content": "Explain x402 in one line."}],
)
print(resp.choices[0].message.content)

Instead of paying per call, you run on a prepaid token balance. Top it up with USDC from your wallet, then spend it across every model — $1 = 1,000,000 tokens, flat across the catalog. No new SDK to learn, no per-call signing, streaming supported out of the box.

One layer underneath, two front doors on top: crypto-native pay-per-use for machines, drop-in compatibility for humans.

Open frontier models, served on NVIDIA NIM

The open models in the marketplace run on NVIDIA NIM — production-optimized inference for frontier open models including NVIDIA Nemotron, DeepSeek, Kimi, and GLM. You get the performance of NVIDIA-optimized serving with the billing model of a marketplace.

Settled on-chain, by the call

Whether you pay per request or draw down a prepaid balance, every dollar is settled in USDC on Solana — fast, final, and verifiable. No invoices, no chargebacks, no waiting on a bank.

Launch offer: 5M free tokens for the first 25 developers

We're opening the OpenAI-compatible API with a launch offer: the first 25 developers claim 5,000,000 free tokens. Connect your wallet, open the dashboard, and claim — no payment required.

👉 Open the Inference Marketplace

This is one layer of a bigger picture. Xona is building the full economic infrastructure for agentic commerce — and inference is where the agents spend.