---
title: "Pay for the answers you ask for, and nothing else"
description: "Each answer costs what the AI provider charged for it, times a published markup. No subscription. Errors are never charged. See a poll's cost before it runs."
canonical: https://mindpolling.com/pricing
updated: 2026-11-19
---

# Pay for the answers you ask for, and nothing else

No subscription, no seats, no plan to choose. Each answer costs what the AI provider charged us for it, times 2.5, and at least $0.001. You add funds in advance, and each run draws on them.

## What an answer costs

The provider's charge for an answer is its tokens (the words in and out, at that model's rates) plus any web searches it ran. We multiply that by 2.5. The markup pays for the card fees, the calls we pay for and never charge (errors, and the measuring every model goes through before it is offered) and running the service.

Every model's rates are on [Models](/models), so any charge can be checked against the answer it came from. Each poll's answers page shows every answer's own tokens and charge.

## What is and is not charged

- **Errors are never charged.** A call that failed is not an answer.
- **A cut-off answer is charged.** The provider charged for it, and you chose the model. An answer cut off at its length limit is asked once more with twice the room, and both tries are charged as one answer, because both were billed.
- **Refused and unreadable replies are charged.** The model replied, and that reply is part of the result.
- **A flex or batched answer is charged at that lower price.** OpenAI and Google models are always asked at their flex tier, and a poll can choose to have its scheduled runs answered through the batch API Anthropic offers. Both charge less for tokens (a search costs the same), so those answers cost less, times the same markup, and never less than the minimum. A request either one does not answer in time is asked again at once and charged at the full price ([flex](/methodology#answered-at-a-provider-s-flex-tier), [batch](/methodology#answered-through-the-providers-batch-apis)).
- **Nothing is charged** for writing, editing or reading a poll, or for downloading its data.

## Before, during and after a run

1. **Before you start**, the poll builder shows a typical and a high estimate for one run and for the whole poll, from each model's measured costs and your own question's length. The high estimate is a guide, not a ceiling.
2. **When a run starts**, it sets aside the most it could cost from your balance.
3. **When it ends**, it is charged what its answers actually cost, and the rest comes back to your balance.
4. **If your balance cannot cover a run**, that run is skipped, never run on credit, and the poll pauses until you add funds. Then it carries on by itself.

## Count the calls first

One question, three wordings, five samples each and three models is 45 answers a run. Daily for 30 days, that is 1,350 answers. Today a typical short answer, with search off, costs from $0.001 to $0.011, depending on the model. So the same poll can cost very different amounts depending on the models and whether search is on. The builder prices your exact choice before you save it.

## Adding funds

Add funds by card, in whole dollars, from $25. The minimum is there because a card payment costs 2.9% plus 30 cents ([Stripe pricing](https://stripe.com/pricing)), which is 5.9% of a $10 payment and 4.1% of $25.

Questions about your balance or a charge: [contact us](/contact).

## Questions

**Do prices change?** A model's price follows its provider's rates. When a provider changes them, the new rate applies to answers from then on.

**Is there a discount for researchers or nonprofits?** No. One price list for everyone.

**What does a poll cost if I stop it early?** Only the runs that ran. Pausing or ending a poll charges nothing.
