Purser
Models at cost

Every AI model at the provider's price. Paid from the agent's own account.

Models are where agents spend the most. On Purser an agent pays for each call at the price the provider charges, from an account its owner controls, with the same rules, firewall and receipts as every other payment.

  • Paid from the agent's account

    Each call is charged at the provider's price and paid from the agent's own account, through its rules and the firewall.

  • Paid before the call

    A small prepaid credit, topped up from the account by at least $0.25 at a time. Each call takes its exact cost from it; what is left stays for the next calls.

  • The cheapest model that does the job

    Ask for fast, balanced or best. Purser picks the lowest priced model at that level that is still inside your budgets.

The price list

Dollars per million tokens, the provider's own price
ModelLevelInputOutput1,000 in, 500 outMarkup
Claude Haiku 5.5Sorting, extracting, routing, short answersFast$0.10$0.50$0.00035$0
Claude Sonnet 5.5Writing, research, most agent workBalanced$2$10$0.00700$0
Claude Opus 5.5Hard reasoning, long plans, codeBest$4$20$0.0140$0

Claude Haiku 5.5 has a second rate for prompts above 100,000 tokens ($0.50 in, $2.50 out), the same as the provider charges. The model price has no markup; when the calls are paid, the usual Purser fee of 0.1% applies to that payment, as on every payment.

How a call is paid

Four steps, every call.

  1. 01

    Paid before the call

    The most the call can cost is taken from the credit first. When the credit is short, the account tops it up on chain, under its rules and the firewall. If it cannot, the model is never asked.

  2. 02

    The call

    Purser sends the request to the provider and returns the answer in the OpenAI format your framework already speaks.

  3. 03

    The exact cost

    The tokens the call really used, times the provider's price. No markup, rounded to the millionth of a dollar.

  4. 04

    The rest stays yours

    What the call did not use goes back to the credit. Top-ups are at least $0.25, so gas never costs more than the calls.

Budgets per model

The owner caps what an agent may spend on each model per day and per task. When the cheapest model has used its budget, a request for a level moves to the next model up. A request for that exact model is refused with the reason.

Cost you can see

Every payment is a receipt with the model and the tokens. The books show cost per model and per task, and every call adds to the open price index of AI work on the network page.

import OpenAI from "openai";

// the agent's own Purser key, nothing else to set up
const ai = new OpenAI({ apiKey: process.env.PURSER_API_KEY, baseURL: "https://purseragents.fun/v1" });

const r = await ai.chat.completions.create({
  model: "auto:fast", // or auto:balanced, auto:best, purser/claude-sonnet-5-5
  messages: [{ role: "user", content: "Summarise this report in three lines." }],
});
// r.purser.costUsd is what the call cost, at the provider's price