How to understand Credit Tokens and usage

What you'll do

Credit Tokens are what you spend as your AI team builds and refines your app. This guide explains how Credit Tokens work and how to keep an eye on your usage, so you're never caught by surprise.

Step 1 — Learn what a Credit Token is

Think of Credit Tokens as fuel for the AI. Every time the AI does real work for you, it uses a little.

Bigger jobs, like having your AI team build a full feature or asking for an AI Review, use more than small chats or quick edits. You don't need to do any math. Analytics tracks it all for you.

Step 2 — Know your amount

Every paid plan includes a set amount of Credit Tokens that arrives fresh each month. The free plan works differently: you get your Credit Tokens once, when you sign up.

• Explorer (free): 250 Credit Tokens, one time — a starting balance, not a monthly allowance
• Starter ($19): 250 Credit Tokens a month
• Builder ($39): 550 Credit Tokens a month
• Accelerator ($99): 1,500 Credit Tokens a month
• Enterprise ($249): 5,000 Credit Tokens a month

If you build a lot, a higher plan simply gives you more room each month.

Step 3 — Check your balance in Analytics

Open Analytics from the main menu, then Credit Tokens & people.

You'll see what you've used this billing period and on what, what came back to you, and roughly how long what's left will last at your current pace.

Step 4 — Understand what a consultation costs

First, the part that costs no Credit Tokens. Ask the AI Team is included in your plan. Asking questions, planning, working out what to build next, getting advice on a screen — that uses your monthly Ask allowance, on every plan including the free one, and it never touches your balance. Credit Tokens pay for the team to make something: a consultation, a build, a review, an image.

A consultation is the biggest single thing you can spend Credit Tokens on. It is not a conversation — it is the full piece of work where your departments interview you and write the blueprint your app is built from. Four things decide its price, and you can see every one of them before you start:

• How many departments your plan brings in. This is the one that moves the number most. A Builder consultation puts 13 specialist departments on your project; an Explorer one puts 3. More departments means more people reading your brief, asking you the questions that matter, and writing up their part of the plan.
• The AI you pick. Every model has a rate, from ×0.5 on the cheapest to ×8 on the strongest. It is the same rate on every paid plan — a bigger plan buys cheaper Credit Tokens, not a cheaper model.
• How long it thinks. Quick ×1.1, Standard ×1.5, Thorough ×2, Intensive ×3, Maximum ×4. Standard is what you get if you leave it alone. A few of the strongest models genuinely think much longer at their deepest settings, so those settings cost more on those models (up to ×8.8) — the picker shows the exact price before you choose, and you always get an answer.
• How deep you want to go. Quick, Balanced or Thorough decides how much each department may ask you. You pick this before you begin, and you can always go shallower afterwards — never deeper, because the price was set when you started.

So a Builder running a Guided consultation on Claude Sonnet 5.5, at Standard thinking and Balanced depth:

25 base × 2.5 (Claude Sonnet 5.5) × 1.5 (Standard thinking) × 3.2 (your 13 departments) = 300

• 134 for the written spec = 434 Credit Tokens

The written spec is the document your team hands to the builder at the end. It is added rather than multiplied, because it is one piece of work whatever settings you chose — deeper thinking does not make it longer.

The starting figure is 25 for almost every model. Two of the most expensive ones cost us more to run than their rate alone accounts for, so they start slightly higher — Claude Fable 5.1 at 28 and GPT-6 Astra at 29. Your breakdown on the start page always shows the figure your model actually uses.

Two models — Gemini 3.1 Flash Lite and Claude Haiku 4.5 — have only one speed, so thinking time does not apply to them and they cost the same whatever you choose.

The number you see is a ceiling, not a bill. You are charged for what your team actually does. Skip questions, or stop early, and the difference comes back — you are never charged more than the figure you agreed to. Nothing is taken until you press Begin, and the start page shows the same breakdown you have just read, for the settings you have actually chosen.

Your plan also sets how long the models may think, which is why a setting is sometimes greyed out with the plan that unlocks it: Explorer and Starter up to Standard, Builder up to Thorough, Accelerator up to Intensive, Enterprise up to Maximum. Within that, every thinking time is offered on every model your plan includes, each at its own price.

Step 5 — Watch your usage over time

Spend a moment in Analytics to see your trend. It shows how much you've used this month and whether your pace is going up or down.

When the usage bar gets close to full, that's your cue to slow down, or move up a plan for more Credit Tokens. On a paid plan you can also wait for next month's refresh. On the free Explorer plan there is no refresh to wait for, so topping up or upgrading is the way forward.

Tips

• On a paid plan, Credit Tokens refresh at the start of each billing month, so a fresh batch arrives automatically. Explorer's 250 are a one-time starting balance and do not renew.
• Large requests, like building a full feature or running an AI Review, cost more than quick chats. Plan the big jobs first.
• If you're regularly running low, upgrading your plan is usually easier than rationing Credit Tokens.

What's next

Ready to change plans or see your billing details? Read How to manage your subscription.