Costs in Statistics

Summary

The Costs tab helps explain total credit spend, how it changed, and where it came from. Use credits as the primary financial indicator; tokens explain model workload but do not replace the final spend amount.

How to read the tab

1. Open the Costs tab.
2. In the cost summary, check charges, average spend, and the balance forecast.
3. On the credit spend chart, find the interval where costs increased or decreased.
4. In the channel breakdown and the adjacent agent breakdown, find the main sources of spend.

Key indicators

• credits used during the period;
• average daily spend;
• average spend per active dialog and message;
• current balance;
• forecast for how many days the balance may last at the current rate.

Average values may be fractional. The forecast may be empty when there were no charges during the selected period.

Full AI response cost

1. The AI responses block shows the cost of the full response cycle: model work and related operations.

The summary includes total spend in credits and its monetary equivalent, the number of AI responses and dialogs containing them, average spend per response, and average spend per dialog. The spend card includes only completed full cycles and excludes cancelled generations.

An AI response is counted after the assistant message is saved; this does not confirm delivery to an external channel.

What makes up the spend

The Usage and costs table groups operations by purpose:

• agent response — input, cached, and output tokens;
• image and speech recognition;
• image and audio creation;
• analytics;
• tool calls for automation, delivery, data, knowledge, and integrations.

Cancelled generations appear as a separate row. It shows the number of attempts, charged credits, and monetary equivalent. This spend is excluded from the completed AI response card but remains actual project spend.

Columns show volume, spend, and share. Each share uses the combined completed and cancelled cycle spend. Tool rows also show calls and errors. Tools are not added to the total twice: the table explains spend that has already been counted.

Monetary equivalent and delay

The monetary amount uses the currency and rate that applied when the operation occurred. If several currencies were used or some operations have no matching rate, credits still count in full while the monetary amount may be incomplete.

New data normally appears after around two minutes, or later while a backlog is processed. While older data is being prepared, the block shows an update state and refreshes automatically.

Where to find individual charges

Open credit history to inspect a specific operation, agent, model, token count, or event. Statistics shows the period-level picture; history shows individual spend records.