> ## Documentation Index > Fetch the complete documentation index at: https://docs.guidinghand.ai/llms.txt > Use this file to discover all available pages before exploring further. # Billing > Prepaid, by tokens: every reply and every task is paid from your org's balance for the tokens its model used, at the price of the agent's model. Eager is prepaid. Every model call an agent makes is paid from the org's balance for the tokens it used: each round of a chat reply, each step of a browser or computer task, and each guardrail check that asks the model. There's no subscription and no per-seat fee, and waiting on a person costs nothing. ## Prices The agent's [`model`](https://docs.guidinghand.ai/concepts/agents#model) sets the price of its tokens, in US dollars per million: | `model` | Model | Input | Cached input | Cache write | Output | Typical task | | - | - | - | - | - | - | - | | `claude-haiku-5-5` | Claude Haiku 5.5 | \$0.15 | \$0.015 | \$0.1875 | \$0.75 | about 1¢ | | `claude-sonnet-5-5` | Claude Sonnet 5.5 | \$3 | \$0.15 | \$3.75 | \$15 | about 15¢ | | `claude-opus-5-5` | Claude Opus 5.5 | \$6 | \$0.30 | \$7.50 | \$30 | about 31¢ | | `claude-fable-5-1` | Claude Fable 5.1 | \$15 | \$0.375 | \$18.75 | \$75 | about 71¢ | | `gpt-6-luna` | GPT-6 Luna | \$0.15 | \$0.015 | none | \$0.75 | about 1¢ | | `gpt-6.1-sol` | GPT-6.1 Sol | \$3 | \$0.15 | none | \$15 | about 10¢ | | `gpt-6-astra`, or `null` (Default) | GPT-6 Astra | \$15 | \$1.50 | none | \$75 | about 58¢ | | `gemini-3.8-flash` | Gemini 3.8 Flash | \$2.25 | \$0.225 | none | \$11.25 | about 32¢ | | `gemini-4-argon` | Gemini 4 Argon | \$6 | \$0.30 | none | \$30 | about 86¢ | * Input read from the prompt cache is billed at the cached price. A task sends what it has done so far with each step, and a conversation its messages so far with each round, so most input is cached. * On Claude models, input written to the cache is billed at the cache-write price (1.25x input, as Anthropic bills it). The other models don't charge for it. * Gemini's thinking counts as output. * A call with a long prompt costs more on two models, as their makers bill it: on Claude Haiku 5.5, a call whose input is over 100K tokens is billed at 5x its prices (\$0.75 input, \$3.75 output); on GPT-6 Luna, over 272K, at 2x input and 1.5x output (\$0.30 / \$1.125). A task on Haiku clears its old screenshots before it gets there. * The typical task is a 10-step browser task, from real tasks on Eager. Real tasks vary with their length and what's on screen. [`effort`](https://docs.guidinghand.ai/concepts/agents#effort) has no price of its own: a higher level thinks more, so it uses more tokens. Prices are also on [guidinghand.ai/pricing](https://guidinghand.ai/pricing). ## What's charged when * **A chat reply** is charged once, when it ends (however it ends): the tokens of all its rounds at the model's price, rounded up to a whole cent. The [response](https://docs.guidinghand.ai/concepts/conversations.md#responses)'s `usage` has its tokens, and the tasks it started with what each cost. * **A task** (one a conversation started, or one you started) is charged as it runs: each model call adds to what it owes, and Eager takes it from the balance every few seconds. When it ends, the rest is charged, rounded up to a whole cent once per task. The task's `usage` and `cost_cents` say what it used and what it cost. See [Tasks](https://docs.guidinghand.ai/concepts/tasks.md#billing). * **Waiting** on an answer, an approval or a form uses no tokens, so a slow reply from a person costs nothing. * **Cloud browsers** and their hosting are in the token price: a task on a browser is billed for its tokens like any other. * If an Eager restart ends a task, what it was charged goes back to the balance. ## Balance and free credit The org pays from a prepaid balance in US dollars. A person's first org gets **\$20 of free credit**, once, in its balance. It pays for replies and tasks alike, needs no card, and isn't refundable. Orgs that still had free agent minutes on 2026-10-18 got them as free credit, at 75¢ a minute. It's all in the console under **Usage & billing**. Everyone in the org sees the balance and this month's cost, by agent and by model. Admins also see past top-ups, with receipts. * **Add funds**: owners add \$10 to \$5,000 by card. The card is saved for later top-ups; replace it with **Update card**. * **Auto top-up** is off until an owner turns it on. When the balance falls below an amount you choose (default \$10), Eager charges the saved card the amount you set (default \$50). * **Monthly limit**: the most auto top-up charges in a calendar month (UTC), default \$500. When the next top-up would go past it, only what's left under the limit is charged (if that's at least \$5); after that, auto top-up waits until next month. Funds you add yourself don't count toward it. * If auto top-up can't charge the card (it was declined, or the bank asked to confirm the charge), it stops trying and **Usage & billing** says why. Add funds or update the card to fix it; saving the auto top-up settings also lets it try again. * Unused balance doesn't expire. What you added is refundable on request; free credit isn't. Only owners, signed in to the console, can add funds or change the card and auto top-up; API keys never can. Admins and API keys can change an agent's `model`, which sets the price of its tokens, and its `effort`, which changes how many it uses. Each change is in the [audit log](https://docs.guidinghand.ai/concepts/audit-log.md), and the monthly limit caps what auto top-up can spend. ## When the balance runs out Sending a message or starting a task when the org's balance is empty returns `402 payment_required`. If auto top-up is on, Eager tries it first and carries on when it works. `error.code` says why it didn't: | `code` | Why | Fix | | - | - | - | | `insufficient_balance` | The balance is empty (free credit included), and auto top-up is off or still going through. | Add funds, or turn on auto top-up. | | `auto_topup_limit_reached` | Auto top-up has charged this month's limit. | Add funds, or raise the monthly limit. | | `card_needs_attention` | Auto top-up couldn't charge the card: it was declined, or the bank asked to confirm it. | Add funds or update the card. | If the balance runs out while a task is running, Eager tries auto top-up and lets the task carry on for a short while: 30 seconds, or up to 90 seconds while a top-up is going through. If the top-up works, the task carries on and pays as usual. If not, the task is stopped, the tokens it used in that time are free, and its `stopped` event says why, for example "The balance ran out. Add funds in Usage & billing to keep going." A reply or task that ends on its own is charged for all its tokens, even when that takes the balance a little below zero. The next one then waits until the org adds funds. ## Limits | Limit | Default | | | - | - | - | | Tasks at once | 1 until the org first adds funds, then 10 | Past it, starting a task is `429` with `code: "concurrency_limit"`. A conversation's browser task waits for room instead. | | Cloud browsers open at once | 2 | See [cloud browsers](https://docs.guidinghand.ai/concepts/computers.md#cloud-browsers). | Ask Eager support if you need more. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.