BuildSpark AI
Dismiss

BuildSpark AI: practical AI concepts for builders. New here? Start with Tokens.

Home · Playbooks · Foundations

Track playbook

Foundations playbook

€39 once

The token budget set is on this page: a worked example of per-request and monthly cost, and a pre-launch cost checklist. Context layout and prompt rewrite are not written yet. The other concepts on this track do not have a practice set yet.

The token budget set on this page is readable now. Context layout and prompt rewrite are not written yet. This playbook is not for sale on this site. The email links open a message to support@buildsparkai.com with the tier and the track in the subject. There is no form and no payment.

Token budget

One worked example of what a request costs, then what a month of those requests costs. The app, the traffic, and the token counts are illustrative assumptions. They are not measured usage and not a claim about a product that exists. The prices are EXAMPLE rates copied from one official pricing page.

The example app

A support-reply helper. One model request drafts a reply to one ticket. Assumptions used only for this arithmetic:

EXAMPLE prices

Model gpt-4.1. Standard prices, per 1,000,000 tokens, from the short-context columns. Input $2.00. Cached input $0.50. Output $8.00. Cache writes for gpt-4.1 are blank on that table, so this example adds no cache-write charge.

Source: OpenAI API pricing. Prices checked 2026-10-17, may change. Other models, the batch table, and long-context rates on the same page are different. This example does not use them. Amounts are US dollars, as on that page.

Per request

The three dollar amounts match only because these assumptions were chosen that way. A different token split would not land on three equal charges. The bar is a picture of this example, not a general rule.

The month, under the same assumptions

Left out of this example on purpose: retries, a second model call, images, tools, cache writes, and any higher rate past a size threshold. If the prefix is not actually billed as cached input, replace the $0.50 rate with the $2.00 input rate and run the same multiplication.

Pre-launch cost checklist

One pass before you ship a feature that calls a model. Each line is something you can do with a request you already send.

  1. List every model call one user action makes, including a retry and a second model if you have one.
  2. On one real request, write down uncached input tokens, cached input tokens, and output tokens as three numbers.
  3. Open the provider's current pricing page. Copy the input, cached-input, and output rates for the model you will ship, and write the date next to them.
  4. Multiply each token count by its own rate. Add the three results. That is the cost of one request.
  5. Write a monthly request count, and label it as an assumption until you have measured traffic.
  6. Multiply the per-request cost by that assumption. That is the monthly estimate.
  7. Check whether a longer prompt crosses a size threshold that uses a higher rate. If it does, price that request on the higher rate.
  8. Use the cached-input rate only after a real request shows those tokens billed as cached input.
  9. If the pricing page lists a cache-write charge, add it. Do not fold it into the cached-input rate.
  10. Choose a monthly number that should trigger an alert, and decide what the product does when that number is hit.
  11. Open the pricing page again on launch day. The EXAMPLE rates above were checked 2026-10-17 and may already differ.

Still to come

The other concepts on this track do not have a named set yet. They are listed below.

Cards this playbook sits beside

These are the concept cards. The cards are free. 9 cards on this track are live. A coming soon name is not a card yet.

Join the waitlist Founding member, about 30% off

See the other tiers on Pricing, or the concept list on the Foundations track.