Tokenomics methodology
How UnitVane AI assigns AI charges to buckets and deployments, and calculates effective cost per 1M tokens.
Pricing buckets
| Bucket | What goes in it |
|---|---|
| Pay-as-you-go | Per-token charges for input, output and cached tokens |
| PTU hourly | Provisioned capacity billed by the hour without a reservation |
| PTU reservation | Reservation charges, with their term: monthly, 1 year or 3 years |
| Other | Fine-tuning, hosting, storage and other AI meters |
Billing correction
Billing exports sometimes record hourly PTU charges as pay-as-you-go, and leave reservation terms blank. UnitVane AI reads billing detail for subscriptions with PTU, moves those charges to the right bucket, and sets the term, in place. Nothing is deleted and daily totals do not change.
Allocation
- Pay-as-you-go cost is assigned to the deployment and model that incurred it.
- PTU hourly and reservation cost is allocated to deployments in proportion to the tokens each served on that capacity.
- Model cost inside a shared deployment is split by each model's token share.
Effective cost per 1M tokens
Allocated cost divided by total tokens, times one million, per model, deployment, team or cloud. Pay-as-you-go prices are read from provider price lists and refreshed daily, so savings estimates use current rates.
Month close
Providers restate charges after a month ends. UnitVane AI reloads the previous month at close, and flags meters it cannot match to a known model or bucket so they can be reviewed.