Lower provider prices are useful, but they are not the number a user actually needs before they run an AI request.
If a model catalog only shows raw provider price, the buyer still has to guess the real settlement price: which route will be used, which balance will pay, whether a fallback can move the request to another channel, and whether the displayed price is official direct access or a discounted route.
That is the pricing problem Tokens Forge is trying to make boring and visible.
Provider price is only the starting point
A provider price answers one narrow question: what does the upstream model usually cost per token?
A settlement price answers the operational question:
- what will this request cost me inside this product?
- will it use official model Credit or a lower-cost RMB route?
- which catalog model did I pick?
- which upstream model will actually receive the request?
- what happens if the primary route fails and a backup route is used?
Those details matter because cheap AI tokens are usually cheap because routing, pooling, credits, channel priority, or margin rules sit between the user and the upstream provider. Hiding that layer makes the product look cheaper in the catalog but harder to trust after the bill arrives.
Model marketplaces need two price columns
For AI model marketplaces, I think there are two prices worth showing clearly:
- Official/provider reference price.
- Settlement price for the route the user can actually run.
The first column is useful for comparison. The second column is what the user cares about before creating an API key or starting a heavy workflow.
In Tokens Forge, official direct models use Credit semantics. Standard lower-cost routes use a separate RMB wallet semantic. The exact labels are product-specific, but the principle is general: do not make users infer settlement behavior from symbols, exchange rates, or hidden routing rules.
Why this matters more for AI research tasks
A short chat completion may be cheap enough that nobody checks the receipt. A longer AI researcher run is different.
A trading research report can call multiple models, fetch data, retry sections, expand context, and produce a longer final report. If the UI only says the provider price, users still do not know the run budget. They need balance warnings, route visibility, and final ledger entries that explain which balance paid for the work.
That is why Tokens Forge keeps the AI Researcher connected to usage ledgers instead of treating it as a separate chatbot feature. The free research workflow is useful only if the token burn stays explainable.
The receipt should match the catalog
After the request runs, the usage receipt should match what the user saw before the request. It should preserve:
- requested catalog model
- upstream model
- route type
- primary or backup channel
- balance bucket
- input and output units
- settlement price used for billing
- retry or fallback context
This is not just finance hygiene. It is product trust. Users buy cheaper AI tokens because they want lower spend, but they stay only if the system can explain the spend after routing decisions happen.
Cheap access is the acquisition hook. Settlement clarity is what makes the product usable for repeated work.
Top comments (0)