Cost statement 2026-10
2026-10-01 to 2026-10-31. Version 1, closed 2026-11-02T10:00:00.000Z by sample generator.
This is cost evidence for AI and AI agents only. It supports an audit; it is not an audited financial statement.
- This is cost evidence for AI and AI agents only. It supports an audit; it is not an audited financial statement.
- The provider-reported total is partial: OpenAI (not_connected), Anthropic (not_connected), Kimi (Moonshot AI) (not_connected), OpenRouter (not_connected), Gemini (Google) (not_connected), GLM (Zhipu AI, BigModel) (no_report_from_provider), DeepSeek (no_report_from_provider), AWS Bedrock (not_connected), Google Vertex AI (not_connected), Azure OpenAI (not_connected), LiteLLM and other gateways (not_connected).
Estimated cost 7360 USD over 67450 calls (0 with no price).
1. Totals by provider and model (estimated)
| Provider | Model | Calls | Estimated USD |
|---|
| sample-ai | sample-large | 22878 | 6368.9003 |
| sample-ai | sample-small | 44572 | 991.0997 |
1. Provider-reported (the provider's own figure)
2. Reconciliation
| Provider | Explained share | Untracked share | Unpriced share |
|---|
3. Allocation by department (estimated)
| Department | Calls | USD |
|---|
| Platform | 19997 | 2944 |
| Support | 27115 | 2208 |
| Finance | 13558 | 1472 |
| (unallocated) | 6780 | 736 |
3. Allocation by cost center (estimated)
| Cost center | Calls | USD |
|---|
| CC-210 | 27115 | 2208 |
| CC-310 | 10958 | 1900 |
| CC-120 | 13558 | 1472 |
| CC-330 | 9039 | 1044 |
| (unallocated) | 6780 | 736 |
3. Allocation by product (estimated)
| Product | Calls | USD |
|---|
| Support desk | 27115 | 2208 |
| Infrastructure | 10958 | 1900 |
| Finance automation | 13558 | 1472 |
| Research | 9039 | 1044 |
| (unallocated) | 6780 | 736 |
4. Budgets
| Scope | Subject | Spend | Monthly cap |
|---|
5. Avoidable cost (estimates)
| Line | Calls | Estimated USD |
|---|
| Calls that failed | 2117 | 244.312 |
| Runs that ended in failure | 2079 | 220.0434 |
| Retries | 822 | 98.8866 |
| Calls repeated in a row | 0 | 0 |
| Loops | 0 | 0 |
| Runs cut off by a limit | 0 | 0 |
6. Incidents with a cost
| Incident | Status | USD at close |
|---|
7. Coverage
| Source | Days with records | Of days |
|---|
| otel:sample | 31 | 31 |
8. Method
- A record belongs to the period of the UTC day it was ingested.
- tokens_in is all input including the cached part; tokens_out is all output including reasoning.
- Estimated savings are projections. Checked savings are read from the evidence after a window, per unit. The two are never added together.
- By the resolved tag on each record (a tag on the record, then on the agent, then a person's department, then a rule). A call with no value is its own "(unallocated)" line and is never spread over the others.
- Each model call is priced when it arrives, from the tokens the source reported and the price table row in effect that day: uncached input, cached read, cache write and output at their own prices, times any batch, priority or flex multiplier. A call with no price is counted and has no estimate.
- Provantix's tokens and cost for each provider, model and UTC day against the provider's own figures, for days that have settled, within the tolerances below.
9. What this statement does not show
- Anything outside AI and AI agents: this is not a statement of the organization's costs.
- What the provider billed or what was paid. The provider's figures are what its own report said at the time they were fetched; an invoice is not read.
- Spend by a provider Provantix is not connected to, or by a source that sent nothing, or calls that were never recorded: those are "untracked" or "not connected" and are named where known.
- Whether any spend was necessary, correct or worth it, or that any call was authorised. Cost per outcome is a separate report and puts no value on an outcome.
- A cause for any difference between Provantix's figures and a provider's.
- That the figures are audited. It is cost evidence that supports an audit; it is not an audited financial statement.