Models & Units

FluxA AgentMarket routes LLM calls to many models (Claude, GPT, Gemini, DeepSeek, GLM, Kimi, Qwen and more) behind one OpenAI-compatible endpoint. There are no provider accounts to open: every call is paid from one prepaid Units balance on your AgentMarket account.

Base URL: https://router.fluxapay.xyz/llm/v1
Wire formats: OpenAI Chat Completions (/chat/completions) and Anthropic Messages (/messages)
Auth: Authorization: Bearer <fxa_live_ API key> or Authorization: Bearer <Agent VC>. See API Keys & Auth
Billing: per token, in Units. 1 Unit = $0.00001: 1 USDC or 1 Monetize Credit buys 100,000 Units

Quick example

bash

curl https://router.fluxapay.xyz/llm/v1/chat/completions \
  -H "Authorization: Bearer fxa_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-opus-5",
    "messages": [{ "role": "user", "content": "Hello" }]
  }'

Any OpenAI SDK works by setting the base URL:

python

from openai import OpenAI

client = OpenAI(base_url="https://router.fluxapay.xyz/llm/v1", api_key="fxa_live_...")
resp = client.chat.completions.create(
    model="anthropic/claude-opus-5",
    messages=[{"role": "user", "content": "Hello"}],
)

No API key? An agent can authenticate with a short-lived Agent VC minted from its wallet identity instead:

bash

VC=$(fluxa-wallet agent-vc --audience fluxa-wallet-service \
  --challenge wallet-user-info-lookup --ttl 300 | jq -r .data.vc)

curl https://router.fluxapay.xyz/llm/v1/chat/completions \
  -H "Authorization: Bearer $VC" -H "Content-Type: application/json" \
  -d '{"model":"z-ai/glm-5.3","messages":[{"role":"user","content":"hi"}]}'

Discover models

`GET https://router.fluxapay.xyz/llm/v1/models`

OpenAI-standard model list. No authentication. Use it to fill a model picker or as a health check.

json

{
  "object": "list",
  "data": [
    { "id": "x-ai/grok-4.6", "object": "model", "created": 1787074829, "owned_by": "aihubmix" },
    { "id": "anthropic/claude-opus-5", "object": "model", "created": 1787074829, "owned_by": "aihubmix" }
  ]
}

A model offered by several providers is listed once. owned_by is the first provider that offers it.

GET https://router.fluxapay.xyz/llm/{provider}/v1/models lists one provider's models. An unknown provider returns 404 with code: "provider_not_found".

Models with prices

GET https://agentmarket.fluxapay.xyz/api/discover?type=model returns each (provider, model) offering with its price:

json

{
  "models": [
    {
      "provider": "aihubmix",
      "id": "anthropic/claude-opus-5",
      "displayName": "Claude Opus 5",
      "categories": ["text", "vision", "thinking"],
      "inputUnitsPer1M": 500000,
      "outputUnitsPer1M": 2500000,
      "interfaces": ["chat", "messages"],
      "source": "fluxa",
      "urls": { "merchantGuide": "https://agentmarket.fluxapay.xyz/marketplace/models/aihubmix/skills.md" }
    }
  ]
}

Field	Description
`provider`	The provider that serves this offering. Use it to pin a provider (see Routing)
`id`	Pass as `model` in the request body
`inputUnitsPer1M` / `outputUnitsPer1M`	Price in Units per 1M tokens. `500000` Units = $5
`interfaces`	`chat` = `/chat/completions`, `messages` = Anthropic `/messages`
`urls.merchantGuide`	Agent-readable guide for that provider

Use search or query to filter, for example ?type=model&query=aihubmix or ?type=model&search=claude. See Discovery.

For full rate cards, including cached-input and long-context tiers, use GET https://agentmarket.fluxapay.xyz/api/llm/models?vendor={provider}. Read the rates_units_per_mtok object of each model (input_tokens, output_tokens, cache_hit, and tiers[] with an above_tokens threshold). GET https://agentmarket.fluxapay.xyz/api/llm/vendors lists providers with slug, name, description, modelCount and priceRange.

CLI: fluxa-wallet market search --models ["<provider>"], fluxa-wallet market search --vendors.

Endpoints

Method	URL	Format
`POST`	`https://router.fluxapay.xyz/llm/v1/chat/completions`	OpenAI Chat Completions. Provider chosen for you
`POST`	`https://router.fluxapay.xyz/llm/v1/messages`	Anthropic Messages. Claude models only
`POST`	`https://router.fluxapay.xyz/llm/{provider}/v1/chat/completions`	OpenAI Chat Completions on one provider
`POST`	`https://router.fluxapay.xyz/llm/{provider}/v1/messages`	Anthropic Messages on one provider

The request and response bodies are the standard OpenAI or Anthropic ones. "stream": true is supported where the model supports streaming.

On /messages, Anthropic clients may send the credential as x-api-key: <fxa_live_ key or Agent VC> instead of Authorization. Errors on /messages come back in Anthropic's error shape with the same status codes.

Routing

On /llm/v1/... the model field decides where the call goes. When several providers offer the model, the one with the lowest input price is used. To pin a provider, either put it in the path (/llm/{provider}/v1/...) or add a provider field to the body:

json

{ "model": "anthropic/claude-opus-5", "provider": "aihubmix", "messages": [ ... ] }

The provider field is removed before the request is forwarded.

Response headers

Header	Description
`X-LLM-Cost-Credits`	What this call cost, in Units
`X-LLM-Balance`	Your Units balance after this call
`X-LLM-Cost-Clipped`	`true` if the charge was capped below the computed cost
`X-FluxA-Request-Id`	Request ID. Quote it when you contact support

Errors

Errors on /chat/completions are JSON with an error field, and often a message:

json

{ "error": "rate_limited", "message": "The model provider is rate limiting requests. Please retry after a short delay." }

Status	`error`	Meaning
`400`	(message)	Missing or invalid JSON body, missing `model`, or a model that is not available
`400`	(message), `code: "interface_not_supported"`	The model is not served on this format. `interfaces` lists the formats it is served on
`400`	`invalid_request`	The model rejected the request (for example, the input is too long). `message` explains why
`401`	(message)	Missing, invalid or revoked credential, or an expired Agent VC. Mint a fresh VC and retry
`402`	`settle_required`	Your Units balance is negative. See Negative balance
`403`	`spend_cap_exceeded`	This API key reached its spend cap. Includes `cap_credits` and `spent_credits` in Units
`403`	`Account suspended`	The account cannot make calls
`404`	(message)	Unknown provider, or no provider serves this model
`429`	`rate_limited`	Too many requests. Honor `Retry-After` if present
`502`	(message)	The model could not be reached. Not charged
`503`	`upstream_unavailable`	The model is temporarily unavailable. Not charged. Retry later

Failed calls are not charged.

Units balance

Units are one balance per account, spendable on any model at any provider. They do not expire.

`GET https://router.fluxapay.xyz/llm/wallet/balances`

Auth: API key or Agent VC.

json

{
  "balance": {
    "balance": 482150,
    "balanceUsd": 4.8215,
    "lifetimeSpent": 17850,
    "burn7dPerDay": 2550
  }
}

Field	Description
`balance`	Units left. Negative means you owe Units (see below)
`balanceUsd`	The same in USD
`lifetimeSpent`	Units spent on model calls, all time
`burn7dPerDay`	Average Units spent per day over the last 7 days

balance is null if the account has never held Units.

CLI: fluxa-wallet market model remainingUsage

`GET https://router.fluxapay.xyz/llm/wallet/ledger`

Top-ups and spending, newest first. Auth: API key or Agent VC.

Param	Default	Description
`kind`	`all`	`all`, `purchases` (Units added) or `spend` (Units used)
`limit`	`20`	1 to 100
`offset`	`0`	Pagination offset

json

{
  "kind": "all",
  "total": 42,
  "offset": 0,
  "limit": 20,
  "entries": [
    { "ts": "2026-10-27T09:12:44Z", "type": "meter", "amount": -1250, "balanceAfter": 482150 },
    { "ts": "2026-10-26T18:01:02Z", "type": "topup", "amount": 500000, "balanceAfter": 483400 }
  ]
}

`type`	Meaning
`topup`	Units added by a purchase
`meter`	A model call
any other value	An account adjustment or refund. Use the sign of `amount`

Entries may carry extra fields. Rely only on the fields above.

CLI: fluxa-wallet market model usageHistory

Negative balance

A call that costs more than your remaining balance still completes, and your balance can go below zero. Once the amount owed grows past a small limit, model calls return 402 with error: "settle_required" and x402 accepts that either clear the exact amount owed or buy a bundle. Pay one of them with FluxA Agent Wallet and resend the call with X-Payment, or simply top up and retry.

Top up Units

Units are sold in fixed bundles. Every bundle grants Units at the same rate: 1 USD = 100,000 Units.

Bundle	Price	Units
`mini`	$1	100,000
`starter`	$5	500,000
`mid`	$10	1,000,000
`pro`	$25	2,500,000

There are three ways to pay:

Rail	Pays with	Best for
x402 (agent)	Monetize Credits or USDC on Base, from the agent's wallet under a spending mandate	Agents. No browser step after the mandate is approved
Payment link	Monetize Credits, through a FluxA Wallet checkout page	A human approves in the browser
Card	Visa/Mastercard via the wallet's linked card	Humans without USDC or credits

Never top up without the user's explicit approval of the bundle and the rail.

Agent top-up with x402

The CLI does the whole flow:

bash

fluxa-wallet market model topup --bundle starter --credit   # pay with Monetize Credits (default)
fluxa-wallet market model topup --bundle starter --usdc     # pay with USDC on Base
fluxa-wallet market model topup --bundle starter --mandate mand_xxx   # reuse a signed mandate

Without --mandate, the CLI creates a spending mandate and prints an approval link for the human. Full agent guide: agentmarket.fluxapay.xyz/marketplace/models/agent-topup.md.

The underlying API has two steps.

`POST https://router.fluxapay.xyz/llm/topup/initiate`

Auth: API key or Agent VC.

json

{ "packageSlug": "starter" }

Answers 402 on success, with the x402 challenge for the order:

json

{
  "x402Version": 1,
  "error": "X-PAYMENT header is required",
  "accepts": [
    { "scheme": "exact", "network": "fluxa-monetize-credits", "...": "..." },
    { "scheme": "exact", "network": "base", "...": "..." }
  ],
  "orderId": "4f6c...",
  "resource": "https://router.fluxapay.xyz/llm/topup/finalize/4f6c...",
  "costCredits": 5,
  "creditsToGrant": 500000
}

costCredits is the price in Monetize Credits (= USD). creditsToGrant is the Units you get. The base accept is present when USDC top-ups are offered.

Status	Meaning
`400`	Body is not JSON, or no `packageSlug`
`401` / `503`	Credential problem / sign-in service temporarily unavailable
`404`	Unknown bundle

`POST {resource}` (finalize)

Pay the challenge with fluxa-wallet x402-v3 --mandate <id> --payload <402 body>, then POST the returned token to the resource URL. No bearer token is needed here.

bash

curl -X POST "https://router.fluxapay.xyz/llm/topup/finalize/4f6c..." -H "X-Payment: <xPaymentB64>"

json

{ "success": true, "replayed": false, "creditsAdded": 500000, "balance": 982150 }

Finalizing a completed order again returns the same result with replayed: true.

Status	Meaning
`400`	Missing `X-Payment`
`402`	Payment rejected or not completed. Not charged
`404`	Unknown order
`409`	The order failed. Create a new top-up

Payment link

`POST https://router.fluxapay.xyz/llm/topup/paylink`

Creates a FluxA Wallet checkout link for a bundle. Nothing is charged until a human opens the link and approves it. Auth: API key or Agent VC.

json

{ "packageSlug": "starter" }

json

{
  "orderId": "8a1d...",
  "creditsToGrant": 500000,
  "amount": 5,
  "currency": "FLUXA_MONETIZE_CREDITS",
  "checkoutUrl": "https://...",
  "statusUrl": "https://router.fluxapay.xyz/llm/topup/order/8a1d..."
}

Status	Meaning
`400`	Body is not JSON, or no `packageSlug`
`404`	Unknown bundle
`502`	The payment link could not be created. Try again

Give checkoutUrl to the human, then poll statusUrl.

`GET https://router.fluxapay.xyz/llm/topup/order/{orderId}`

Auth: the same account that created the order.

json

{ "status": "settled", "creditsToGrant": 500000, "balance": 982150 }

`status`	Meaning
`pending`	Processing. Waiting for payment
`settled`	Completed. The Units are on your balance
`failed`	Failed. Create a new order

Treat any other value as processing. Poll every few seconds; do not loop forever.

Card

An agent can pay with the human's linked card, through FluxA Agent Wallet, without ever handling a card number. The agent creates an order with POST https://router.fluxapay.xyz/pay/createOrder/topup/{bundle}, the human approves a card spending limit once, the wallet pays, and the agent polls the order's statusUrl until status is settled. Follow the published step-by-step guide: agentmarket.fluxapay.xyz/marketplace/models/topup.md.

Card orders also appear as scheme: "card" entries in a settle_required 402.

Humans can also top up on the website at agentmarket.fluxapay.xyz/marketplace/models.

Models & Units ​

Quick example ​

Discover models ​

GET https://router.fluxapay.xyz/llm/v1/models ​

Models with prices ​

Endpoints ​

Routing ​

Response headers ​

Errors ​

Units balance ​

GET https://router.fluxapay.xyz/llm/wallet/balances ​

GET https://router.fluxapay.xyz/llm/wallet/ledger ​

Negative balance ​

Top up Units ​

Agent top-up with x402 ​

POST https://router.fluxapay.xyz/llm/topup/initiate ​

POST {resource} (finalize) ​

Payment link ​

POST https://router.fluxapay.xyz/llm/topup/paylink ​

GET https://router.fluxapay.xyz/llm/topup/order/{orderId} ​

Card ​

Models & Units

Quick example

Discover models

`GET https://router.fluxapay.xyz/llm/v1/models`

Models with prices

Endpoints

Routing

Response headers

Errors

Units balance

`GET https://router.fluxapay.xyz/llm/wallet/balances`

`GET https://router.fluxapay.xyz/llm/wallet/ledger`

Negative balance

Top up Units

Agent top-up with x402

`POST https://router.fluxapay.xyz/llm/topup/initiate`

`POST {resource}` (finalize)

Payment link

`POST https://router.fluxapay.xyz/llm/topup/paylink`

`GET https://router.fluxapay.xyz/llm/topup/order/{orderId}`

Card