Key Takeaways

• A credit carries no fixed value across products. Figma charges 20 credits per Add interactions run, monday.com 8 per AI block action and 120 per meeting hour of AI Notetaker.

• Rates run flat per action, per model, or per complexity band. monday vibe prices Gemini Flash at 10 to 20 credits a message, Claude Opus at 50 to 500.

• The rate isn't a promise. Figma gave Professional plans 2x the add-on credits at one unchanged price on 25 August 2026.

• Credits usually aren't money. Figma states they "do not represent legal tender, currency, or stored value, and are not redeemable for monetary value."

What does one AI credit represent?

One AI credit represents whatever the issuing vendor says it represents. No industry unit sits behind the word, no cross-vendor exchange rate exists, and on most products no fixed link to tokens or dollars either. The credit lets a vendor price an action instead of a model call, then reprice it later.

Published rates show how far apart the same word lands:

• Figma bills image work per image: 1 to 5 credits to remove a background, 5 to 10 to boost resolution. One generated image costs 2 credits on ChatGPT Images 2.0 and 16 on Nano Banana 2.

• monday.com bills AI blocks at 8 credits per action, counting repeats on one item inside 24 hours once, and AI Notetaker at 120 credits per meeting hour.

• monday.com bills agents in bands, from 10 to 50 credits for a simple task up to 250+ for an extra complex one.

Credit burn-down covers how the balance drains and which grant goes first.

How do vendors set the credit-to-usage conversion?

Vendors take their own cost for an action, add margin, and round to a number that fits in a UI. Figma names its inputs outright: the model's cost to Figma, the value the feature delivers, and observed consumption. It's a pricing decision dressed as a unit of measure, so the rate moves when any input moves.

Two details matter more to me than the numbers. Canva's 200-versus-20 split is a 10:1 exchange rate published without naming the unit, and monday.com's per-model spread means one balance buys ten times less work on Opus than Flash.

Where do AI credits go wrong for customers?

Credits go wrong when nobody can predict what a task costs before running it. Figma says that because the model decides which actions to take, it can't predict consumption in advance, and shows the number only afterwards.

The failure modes that surface in vendor help centres:

• Repricing under the customer. Figma states consumption changes when the underlying model or its cost changes, so one prompt can cost more next quarter with no plan change.

• No refund for bad output. Undoing an AI action in Figma reverts the file and keeps the credits, because the task already ran.

• Expiry nobody reads. Figma credits never roll over, and Canva top-ups expire 30 days after purchase. Credit rollover has the carry-over rules.

• No cash value. Figma's terms rule out stored value and cash redemption, so an unused balance is a customer write-off.

• Throttling in place of a stop. Canva slows generations rather than blocking them, so an exhausted allowance reads as a slow product.

• Warnings that arrive too late. monday.com notifies at 80% and 100% of the account limit and runs some capabilities past it so an admin can top up.

Should you bill in credits or in tokens?

Bill in credits when your product spans several AI actions with different cost profiles, and bill in tokens when your customers reason in tokens and want the pass-through.

• Credits hide model swaps, put cheap and expensive actions on one balance, and let non-technical buyers compare plans. They also invite the charge that you're marking up compute behind an opaque unit.

• Tokens map to a figure the customer can check against provider pricing. Token-based pricing breaks down the metering that unit demands.

• Outcomes skip both. Intercom charges $0.99 per Fin resolution and nothing for a failed attempt, moving the cost of a bad answer onto the vendor.

Whichever unit you sell, the plumbing is the same: meter the event, price it, deduct, then stop or charge on exhaustion. Credits and Wallets documents how grant priority, expiry, and top-up thresholds get configured.

Related terms

Choosing credits as your unit forces a decision on each of these, and every one has its own page.

• Credit burn-down turns a published credit rate into a balance the customer watches fall.

• Credit rollover answers whether an unused credit survives the month that issued it.

• Token-based pricing is the alternative unit, priced against what the model consumed.

• AI token pricing covers what those tokens cost before a credit wraps them.

• Spending cap caps exposure for buyers who won't accept a credit balance at all.

• AI pricing models sets credits beside seats, usage, and outcome pricing.

FAQ

Can I get AI credits refunded when the output is wrong?

Almost never, because the vendor already paid for the inference. Figma refunds nothing on undo, since the task completed. Intercom is the exception, charging nothing for failed attempts.

Why did the same prompt cost more credits the second time?

Because agentic features price the work the model chose to do, not the words you typed. Figma names the model selected, the task's complexity, and the context handed over, chat history included. A long conversation costs more than a fresh one.

Is one AI credit the same on every product?

No. With no shared definition, a credit can buy a fraction of an image edit on one platform and a fraction of an agent run on another. Comparing plans by credit count tells you nothing until you convert both into actions you'd run.