What things cost
Credit rates per model, the ceiling on a single generation, and the hosting bills that never pass through MaShop.
Credits
Three things in MaShop cost credits: a message to the assistant, a content generation, and an automation run. Everything else is either free or billed by someone other than MaShop.
Building
A message is priced on the tokens it actually consumed, at the rate of the model that answered. Tokens are the units a model reads and writes. A long conversation reads more tokens than a short one, because the whole history is sent again each time.
Three models can answer you, and Auto picks between them. These are the catalogue rates, read in production on 1 September 2026. One credit is one cent.
| Model | Per 1 000 tokens read | Per 1 000 tokens written |
|---|---|---|
| GLM 5.2 | 0.70 credits | 2.20 credits |
| Kimi K3 | 1.50 credits | 7.50 credits |
| MiniMax M3 | 0.15 credits | 0.60 credits |
A worked example. A message on GLM 5.2 that reads 20 000 tokens of context and writes 4 000 tokens of code costs 14.00 plus 8.80, so 22.80 credits. That is 23 cents.
The model you pick therefore changes the bill by a factor of ten. Kimi K3 writes at more than three times the price of GLM 5.2, and twelve times the price of MiniMax M3.
A rate can be changed in the catalogue at any time. The badge under each reply is what was really debited, and it wins over this table.
Why no price is shown before you send
MaShop cannot quote a message in advance, and does not pretend to. The final price depends on the model that actually ran, and the Auto router may change that for you. It also depends on the tokens consumed, which are only known at the end. A message carrying an image is always routed to the model that can read images, whatever you picked.
So MaShop reserves an estimate before the call, then corrects it on the real consumption. If the run cost less, the difference goes back to the bucket it came from. If it cost more, the extra is taken in the same order. Whatever none of your buckets covers is absorbed by MaShop, not pushed onto you.
The ceiling on one generation
A full generation is capped at 3.00 dollars of real provider cost, which is 1 500 credits. If your balance cannot cover that, the ceiling drops to what your balance can actually pay.
Crossing the ceiling does not fail the run outright. The agent is cut, and the work already produced goes through the same build check as any run. If it still does not compile after the repair rounds, nothing is committed and your repository does not move.
Automating
An automation run is billed after it finishes, at its real cost, with the same formula. Nothing is held in advance. The rule is deliberate. An automation runs at six in the morning with nobody watching. A reservation left hanging by a crash would freeze credits you could not recover.
Three consequences worth knowing.
- A run that fails is still billed. It burned tokens across every attempt it made before giving up.
- A run stopped by its time ceiling is not billed at all. The call was still in flight, so the real cost is unknown, and MaShop absorbs it.
- A run that crosses the cost cap you set on that automation is billed, then that automation is suspended. Other automations on the same agent keep running.
A conversation in the Automations chat is billed exactly like a message in the builder. It is the same project and the same credits.
Hosting
MaShop charges nothing to put your work online, because it does not host it.
Publishing creates a Supabase database, a GitHub repository and two Netlify sites on your own accounts. Those three bills are yours, at your own tiers, and never pass through MaShop. The Supabase free tier allows two active projects, and MaShop says so explicitly when you hit that wall.
Deploying and redeploying costs zero credits. The build runs on your Netlify account. What is limited is the count, not the price.
| Plan | Deployments |
|---|---|
| Free | 3 per day |
| Pro and Max | Unlimited, the counter is never touched |
Only a deploy you ask for moves that counter. The first build inside a publish is exempt, and so is the deploy that follows a generation.
Note that the plan card inside the app still reads 3 deploys / 30 days. The rule enforced by the server is three per day.
What a paid plan unlocks
Two actions are refused on the Free plan whatever your credit balance. Publishing a project for the first time, and exporting it. Both return an upgrade prompt rather than a charge.