How Credit-Based Pricing Works, and Why It Is Metered Per Action

Most software charges you for capacity you might use. Seats you might fill, features you might switch on, a tier you might grow into. It is easy to bill and it has almost nothing to do with what you got.

Synton meters work. Credits are the single unit of work in the product: every AI action spends credits from one balance, and every feature module is included in every paid plan. Credits are the only thing you meter, which means the pricing page has one number on it instead of a matrix.

Here is how that actually behaves, including the parts that are easy to get wrong.

Two ways an action is priced

Token-metered work covers chat, agents, and anything where a language model reads and writes. The charge follows the real size of the request and the reply. A typical chat turn lands around 16 credits. The consequence is that two runs of the same agent can cost different amounts, because they did different amounts of work.

Fixed-price work covers things with a known unit cost, such as generating an image or running a competitor scan. These carry a published credit price that you can see before you commit.

What things actually cost

A representative sample, all in credits:

Chat. A short question with a short answer is about 5. A normal turn with the usual context is about 16. A turn where the assistant calls tools and reads the results back is about 28. A deep audit or a large agent job can reach several hundred credits.

Content. Generate an image 12. Edit an image 16. Analyse an image 12. A five-second video clip 14, ten seconds 28, twenty seconds 56. A voiceover 8. A music track 16. Analyse a document or invoice PDF 16. Translate a product 8. Generate written content 20.

Agents and workflows. A light agent task 40, medium 120, heavy 200. A workflow AI step 12, a browser step 20. Sandbox execution 20. A virtual machine, per hour, 60.

Research. A competitor scan 40. SEO analysis 20. A market intelligence lookup 12. An ad optimisation pass 20. A web search 4. A SERP query 2.

Utilities, cheap on purpose. Calculator, data transform or code utility 2. A PDF operation, image operation, invoice, brand asset or contract fill 5. Office document conversion 8.

The store security and health audit is 0. It runs no language model, so you can run it as often as you like. That is a real zero, not a rounding.

One thing to watch: the highest video quality tier costs dramatically more than the standard one, because the upstream render cost is dramatically higher. A five-second clip at the top tier is 1,020 credits against 14 at the standard tier. Check the credit badge on the render button before you queue a batch.

The prices differ because each one tracks the real upstream cost of the work, which is why an image and a search engine query differ by an order of magnitude. When an upstream provider changes its price, the credit price changes with it.

If a price shown on a button ever disagrees with the docs, trust the button. It is read live from the same table the charge comes from.

What is never billed

This is the part that matters most in practice, because it is where usage-based pricing usually goes wrong.

Synton charges after an external call succeeds, not before it. A failed generation, a timed-out scrape, or a vendor error is never billed. Where a charge has to happen before the outcome is known, the charge is reversed, and a reversal is capped at what was actually taken, so it can never create credits out of a refund.

A result served from cache costs nothing. The cache check runs before the charge, so a repeated lookup is not billed twice.

Context that Synton has already cached is not counted again, which is why later turns in a long conversation are usually cheaper than the first.

And when your balance cannot cover an action, Synton refuses it before making any external call. You get a dialog rather than a half-finished job. Nothing is charged and nothing is left half-done. The one exception to be aware of: if you see that dialog partway through a long agent run, the steps that already completed were charged and the remaining steps were not.

For actions with a known price, the number is shown inline before you confirm, as a small badge reading roughly N credits next to the button that starts the work. That number is read from the same table the charge is taken from, so the estimate and the charge cannot drift apart. Chat and other token-metered work cannot show a fixed number in advance, because the real answer depends on what you asked.

The plans

Two surface plans exist: Self Service and Enterprise. Self Service has nine tiers that differ only by how many credits you get each month. Every tier unlocks every feature.

• 10,000 credits a month. $49
• 19,000 credits a month. $89
• 32,500 credits a month. $149
• 45,000 credits a month. $199, the most popular tier
• 70,000 credits a month. $299
• 108,000 credits a month. $449
• 160,000 credits a month. $649
• 250,000 credits a month. $999
• 390,000 credits a month. $1,499

Every Self Service tier includes all the AI agents, Shopify, WooCommerce and Magento, every feature module, Meta and Google Ads, API access, 50% credit rollover and priority support. Nothing is locked behind the tier you started on, so an upgrade changes your monthly credit grant and nothing else about what you can do.

Bigger plans give you more credits, never a cheaper credit. The per-action prices above are identical on every plan and every platform. That is deliberate: it keeps the balance fungible and stops the pricing page from becoming a puzzle.

Yearly billing is 17% less for the same credits, calculated as exactly 17% off twelve monthly payments, rounded to whole dollars.

Enterprise adds SAML SSO, a dedicated success manager, a DPA and custom SLA, and unlimited stores.

Two access notes: Plan and Billing is visible only to the account Owner, and your payment method is entered on a hosted checkout page rather than inside Synton.

There is no free trial

Being direct about this, because it is a real difference from most of the category.

There is no free trial and no signup credit grant. Creating an account is free, takes an email and a password, and needs no card, no sales call and no trial clock. But every action that calls an AI model needs an active paid plan. You will see the plan prompt the first time you ask Synton to do AI work, not before.

What you can do without paying anything: connect your store and explore every app, read your dashboard, store overview and store health reports, run the security and health audit as often as you like, use the Reviews, Bundles and Popups modules, and store up to 250 MB in Drive.

What needs a plan: chat, agents, image and video generation, AI SEO, content generation, AI translations, competitor analysis, the customer-facing chatbot and the ad optimizers.

The reason is unglamorous and worth saying out loud: every AI action has a real cost the moment it runs. A free tier of AI work is a bill someone else pays, and eventually that shows up as either a worse product or a worse price.

How the balance actually behaves

Your balance has two pools, and this is the part people most often misread.

Daily credits are your monthly grant divided by 30. They are spent first, and they are restored on your first billable action each day.

Rollover credits are unused daily credits plus every credit pack you have bought. They are spent second. Both pools are spent in one atomic step, so a single action can straddle the two without double counting.

The rollover cap is 50% of your monthly grant. On the 45,000 tier that means 1,500 daily credits and a rollover ceiling of 22,500.

Now the misreading. Daily credits are allocated on your first billable action of the day, not by a job that runs at midnight. A store that sits idle for a week does not accumulate seven days of allowance. You come back to one day's worth in the daily pool, plus whatever moved into rollover along the way.

Two better pieces of news. The cap never claws back an existing balance: credit packs are granted straight into rollover, and if a pack pushes you above the cap, the cap does not take the difference back. It only stops further plan allowance from being added. And rollover survives renewal, because it is a running balance rather than a per-month bucket. Nothing resets it when your plan renews.

If you cancel, the plan stops granting new daily credits at the end of the paid period. Credits already in your balance are not refundable and have no cash value.

Top-ups

Three packs, bought under Settings, Plan and Billing, Credits:

• Starter, $5. 500 credits
• Plus, $10. 1,200 credits
• Pro, $25. 3,500 credits

The larger packs carry a 20% and 40% bonus respectively over the starter rate, and active subscribers get 20% off every pack, because your subscription already covers the fixed costs. Credits appear on your balance within about a minute of the payment confirming, and each top-up is recorded with a 12-month validity date visible in your top-up history.

Auto top-up will buy a pack when your balance falls below a threshold you choose. One gotcha: the toggle stays disabled until a payment method is saved, and a payment method is only saved by a successful pack purchase, so you have to buy one pack manually first. The balance sweep runs hourly, so a top-up fires shortly after you cross the threshold rather than instantly.

Shopify stores work differently. Synton is a Managed Pricing app on Shopify and cannot create a separate one-time charge, so top-ups are billed as a usage charge against your subscription and appear on your next Shopify invoice. You buy a credit amount rather than a named pack, minimum 1,000 and maximum 75,000 per purchase, and you automatically pay the subscriber rate. It requires an active paid subscription, and Shopify's own monthly usage cap can refuse a purchase.

Credits are non-refundable and cannot be converted back to money.

Email

Email is metered at 10 emails per credit, rounded up, so a send of 51 emails costs 6 credits rather than 5.1. Rounding is per charge, so one campaign to 1,000 recipients costs less than a hundred separate sends of 10.

Each tier also has a monthly email ceiling, which is your monthly credit grant expressed in emails. Treat that as the point where Synton stops you, not as a target: sending 100,000 emails on the smallest tier would use every credit you have.

Two honest caveats. Email metering is being switched on store by store and is off for most stores today, and there is currently no screen that tells you which state you are in. And there is no email quota meter in the product yet. The nearest thing is Settings, Plan and Billing, Usage, and the daily credit burn chart.

Why meter at all

Because the alternative is worse. A flat fee for unlimited AI either gets priced for the heaviest user, which overcharges everyone else, or gets rationed quietly through throttling and queues you cannot see.

Metering per action makes the cost of a decision visible before you make it, keeps the failure cases free, and means a light month costs you less than a heavy one. It also means we have no incentive to hide what an action costs, because the number on the button is the number on the bill.