A 14-day example with three cost buckets
These provider and model labels are fictional. They illustrate the normalized format accepted by ModelMeter, not current provider prices.
| Project / model | Requests | Billed USD | USD / 1,000 requests |
|---|---|---|---|
| Support / Fast model | 12,000 | $48 | $4 |
| Research / Reasoning model | 500 | $86 | $172 |
| Support / Backup model | 800 | $6 | $7.50 |
Total billed spend is $140 over 14 complete days. Research accounts for $86 / $140 = 61.43% of that spend, despite having fewer requests. This identifies a bucket to inspect; it does not establish that the model is wasteful or replaceable.
Monthly pace is an assumption
monthly_pace = billed_spend / complete_days × days_in_month $140 / 14 × 30 = $300 $300 / 40 active_customers = $7.50 per customer
This calculation assumes the remaining days behave like the observed days. A launch, trial cohort or traffic spike can invalidate it. It is neither a forecast nor your next invoice. An entered $300 monthly budget is exactly matched by this example's pace.
Prepare a comparable export
- Keep project, provider, model, requests, input_tokens, output_tokens and billed_usd headers.
- Aggregate to non-overlapping rows for the same complete time window.
- Reconcile the billed USD column with the provider's invoice or dashboard. Resolve currency conversion and credits outside this tool.
- Enter complete days, calendar days in the month and the customer count for your chosen definition.
- Review expensive groups before changing routing, quality thresholds or quotas.
What does this report omit?
Costs absent from your CSV remain absent, including infrastructure, storage, retrieval, tool calls and support. Provider-native exports are not automatically mapped. Input and output tokens are retained as context; ModelMeter does not apply a hidden price table to them. Files are processed locally, with no provider key required.