Tokenpilot

Is my AI API bill right, and which lever cuts it most? Reconstruct a month of OpenAI / Anthropic / Gemini spend on today's published prices, price every saving lever (caching, Batch, smaller model, retries, output caps), and see the date your budget runs out. Your numbers stay in this check — nothing is stored unless you press Share.

Or start from the pre-filled workload below: change the five numbers that matter (provider, model, requests/month, input tokens, output tokens) and press Check my bill. Under two minutes.

Your result will appear here.

Your workloads

Check as of (date)

Optional extras

Last month's actual invoice, USD (to reconcile) Monthly budget, USD Spend so far this period, USD Billing period starts Your FX rate, GBP per 1 USD (your rate, default 0.79) Show a 20% VAT view (UK: reverse charge may apply — check your invoice) Paste usage CSV instead (optional) — download the template. Columns like model, input_tokens, cached_input_tokens, output_tokens, num_requests are matched automatically; unknown models are flagged, never guessed.

What this cannot do

Tokenpilot reconstructs token spend from the workloads you describe. It does not see your provider account, so it cannot price web search and other tool fees, image/audio models, fine-tuning, negotiated discounts or Gemini cache-storage hours. If your invoice is higher than the reconstruction, the report lists the likely causes in order. Prices are the published tables bundled on 11 Nov 2026 — providers change them, so confirm big decisions on the pricing page and your invoice.

Free alternatives: each provider's usage dashboard shows what you spent with that one vendor; simple calculators price one hypothetical workload. Tokenpilot adds multi-workload reconstruction, lever-by-lever savings with caveats, cache break-even versus your call gap, invoice variance, and a budget run-out date with dated next steps.