Calculator

Usage CSV / invoice cost analyzer

Drop in an export with model names and token counts. CentsPerToken matches rows to the catalog, estimates spend, and compares against optional invoice cost columns—without uploading your data.

CSV spend

Catalog estimate from usage rows

$127.65est · invoice $93.36

4/5 rows matched

Matched rows80%
4 rows
Unmatched1
Need model id/name map
Invoice total$93.36
Variance-$34.285
Models in summary5
Unmatched rows1

Upload

Usage CSV

Runs entirely in your browser—nothing is uploaded to a server.

Drop CSV hereor choose a file

Headers like model, input_tokens, output_tokens, optional cached_tokens, requests, cost_usd.

  • 1 row(s) did not match the CentsPerToken catalog—estimates skipped for those lines.

Spend

Bars by model

Estimated catalog spend ranked across matched models.

Claude Sonnet 4.5$78.30
Gemini 2.5 Flash$32.63
GPT-5 mini$10.96
GPT-4o mini$5.76
Claude Sonnet 4.5
3,000 req · 15,000,000 in / 2,100,000 out · claude-sonnet-4-5
$78.30
Gemini 2.5 Flash
15,000 req · 52,500,000 in / 6,750,000 out · gemini-2-5-flash
$32.63
GPT-5 mini
8,000 req · 17,600,000 in / 3,200,000 out · gpt-5-mini
$10.96
GPT-4o mini
12,000 req · 21,600,000 in / 4,200,000 out · gpt-4o-mini
$5.76
?
unknown-model
500 req · 500,000 in / 100,000 out · unmatched

Invoice − estimate = -$34.285. Large gaps often mean negotiated rates, batch discounts, or unmatched rows.

Token columns are treated as per-request figures multiplied by requests. Estimates use CentsPerToken list prices, not your negotiated rates. Updated 2026-07-31. Approximate static list prices for planning only. Always verify against each provider’s official pricing page before production budgeting.

Guide

How to get value from this calculator

Paste usage exports or invoice line items to estimate costs locally and compare against billed totals. Helps reconcile token counts, model mix, and pricing tier without uploading sensitive data to a server. Use after each billing cycle to catch drift between planned and actual spend.

Cost formula

Estimated cost = sum over rows of (input tokens / 1M × input price) + (output tokens / 1M × output price) per model; Variance = invoice total − estimated total

Why it matters

Planned budgets diverge from invoices when model mix shifts, caching kicks in, or retries inflate tokens. Regular CSV reconciliation connects engineering behavior to finance line items.

How to use it

  1. Export usage data from your provider dashboard or billing CSV.
  2. Paste the data into the analyzer in the supported format.
  3. Map columns to model, input tokens, and output tokens if prompted.
  4. Review estimated cost by model and total compared to your invoice.
  5. Investigate large variances—often cached tokens, batch tiers, or price changes.
  6. Feed corrected per-model usage into monthly budget planning for next cycle.

Planning tips

  • Include all projects and API keys—orphaned keys are a common source of surprise spend.
  • Compare input vs output token ratio month over month to spot prompt bloat early.
  • Check for batch vs realtime tier mismatches in your estimation assumptions.
  • Processing stays local in your browser—suitable for sensitive usage data.
  • Use the spend-by-model bars to find which model drives the invoice before deep-diving rows.
  • Archive monthly CSVs to build a usage trend baseline for forecasting.

Frequently asked questions

Is my usage data uploaded to CentsPerToken?

No. CSV analysis runs locally in your browser. Data is not sent to CentsPerToken servers.

Why might estimated cost differ from my invoice?

Common causes: outdated list prices in estimates, credits, enterprise discounts, cached or batch tier pricing, taxes, and rounding. CentsPerToken uses approximate list prices for planning.

Which export formats are supported?

Paste provider usage CSVs with model and token columns. Format requirements are shown in the tool workspace.

Can I use this for multi-provider billing?

Yes. Paste exports from each provider separately or combine if columns are compatible. Verify each provider's pricing tier in your assumptions.