Usage & Cost

What the model has cost you, and what the platform is not sure about.

Time range

Preset windows, or custom dates. The exact interval being shown is printed underneath, in your local time, so a chart is never ambiguous about what it covers.

Summary

Figure	Means
Configured cost	Priced with your prices, taken from the snapshot stored with each invocation
Invocations	How many model calls, split into succeeded and failed
Tokens	Input, output, cache read, and cache write, counted separately
Provider estimate	What the provider itself reported, where it reported anything

Two things worth knowing.

Prices are frozen at the time of use. Each invocation stores the prices that were in effect when it ran, so editing today's prices does not rewrite what last month cost.

The provider estimate is reference only. It is not a billing statement, and the page says so. Your bill comes from your provider.

The warnings under the summary

These are the honest part of the page, and they are shown rather than folded into the total.

Warning	Means
Invocations with token usage but no configured price	Real work happened that the platform cannot price, because you have not entered a price for that model
Invocations that have not reported token usage	The provider returned no figures, or the work is still running

A cost cap will not treat unknown cost as zero. If usage is pending or unpriced, automatic work pauses rather than risking an unbounded spend. That is why an unpriced model is worth fixing: it does not just leave a gap in a chart, it can stop automatic investigations.

flowchart TD
    Call["A model call finishes"] --> Usage{"Did the provider<br/>report token usage?"}
    Usage -->|"No"| Pending["Counted as<br/>usage not reported"]
    Usage -->|"Yes"| Price{"Is there a configured<br/>price for that model?"}
    Price -->|"No"| Unpriced["Counted as<br/>unpriced"]
    Price -->|"Yes"| Cost["Priced with the snapshot<br/>taken at the time"]
    Cost --> Total["Configured cost"]
    Pending --> Cap["A cost cap treats these<br/>as unknown, never as zero"]
    Unpriced --> Cap

Cost over time

Daily configured cost. Red markers are days with failed invocations, so a spend spike caused by retries is visible rather than buried.

Setting caps

Caps are set by a platform administrator under Platform administration → Platform settings, not on this workspace's Settings page. They are rolling 24-hour windows, per workspace and per monitor, by count and by cost. Platform settings lists each cap with its bounds and fallback.