How cost is calculated
Learn where Tokenize's spend numbers come from, what the spend labels mean, and why amounts can show as Unpriced or $4.20+.
Tokenize gets cost from two places: the bill your provider reports, and its own estimate from recorded usage. Every spend figure tells you which one you're looking at.
Where cost comes from
From your provider. When you connect a provider's admin key, Tokenize reads what that provider charged. This applies to:
- Claude, with a Claude Enterprise key. See Connect Anthropic.
- Cursor, with a Team Admin API key. Cursor's figure includes its token fee and usage your plan covers. Seat fees aren't included. See Connect Cursor.
- Codex, with an OpenAI Enterprise key. See "Codex credits" below.
Estimated by Tokenize. Where there's no bill, Tokenize calculates cost from the tokens each request used and the model's published list price. Your provider may charge a different amount, for example if you have a negotiated rate.
A provider bill always wins over an estimate when both exist for the same usage.
Spend labels
Most spend figures have a small colored dot next to them. Hover over it to see where the number came from:
| Label | Meaning |
|---|---|
| From your provider | Charges your provider reported. Your final invoice may include other charges or adjustments. |
| Estimated by Tokenize | Calculated from recorded usage and model prices. |
| Provider costs and estimates | A mix: provider charges where you've connected one, estimates for everything else. |
| Some cost is missing | Only part of the cost is available. The amount shown is lower than the full total. |
| Cost unavailable | No cost data for the dates and filters you chose. The amount shows "—". |
The tooltip also says when the data was last updated and which date the provider's records run through.
Why recent days show "Some cost is missing"
Providers report charges a day or more after the usage happens. Until a provider has reported a day, Tokenize leaves that day's billed cost blank instead of guessing. Ranges that include the last day or two often show Some cost is missing for this reason. The gap fills in once the provider reports.
Unpriced requests
Sometimes Tokenize can't price a request, usually because the model is new and isn't in its price list yet. Those requests count as $0, and the amount is marked:
- Unpriced means none of the requests could be priced.
- $4.20+ means part of the cost is known, and the real total is higher.
- A warning icon next to the amount says how many requests aren't priced.
Requests sent through OpenRouter's own routing models can't be priced from tokens. Their billed cost is in Platform Usage.
Unpriced requests don't affect amounts that come from your provider's bill.
Total Spend and API Bill
The Overview shows two separate totals:
- Total Spend is what your people spent using AI tools like Claude Code, Codex, Cursor, and Claude.
- API Bill is what your providers charged your API keys, from Claude Platform, OpenAI Platform, and OpenRouter admin keys. It usually comes from products and automations calling the API.
Tokenize never adds the API Bill to Total Spend. See See API key spend.
Codex credits
OpenAI meters Codex Enterprise usage in credits. Tokenize turns credits into dollars with a price per credit. It uses OpenAI's list price unless you set your own. Changing the price reprices all Codex spend, including past spend. See Set your Codex credit price.
Without an OpenAI Enterprise key, Codex cost is estimated from tokens like any other tool.