An agent that loops overnight can burn a month of budget before you wake up. This workbook prices every agent before it ships, using the 8 October 2026 rates for GPT-6.1 Sol, Claude Haiku 5.5, Mistral Large 4, Gemini 3.5 Flash and more, with the source and confidence of each number on the sheet.
What You Get
- Price table with sources. Eleven model and tier rows with input, cached, output and cache-write prices, a source, a source-type label and a check date. Third-party figures are marked as such.
- Per-agent budget calculator. Pick a model from a dropdown and enter runs, tokens and cache hit rate. Formulas return cost per run, monthly cost, share of cap and an OK, WARN or OVER CAP status.
- Cache-hit sensitivity table. Monthly cost for five models at hit rates from 0% to 95%, plus the percentage saved, so you see what prompt ordering is worth.
- Break-even sheet against human cost. Compares the agent with a person doing the task, counting review time, redo rate, fixed cost and one-off build cost, and ends with a verdict.
- Alert thresholds and projection. Info, warn, stop and spike thresholds, a month-end projection from spend so far, and a six-step runbook.
- Plans sheet. ChatGPT and Codex plans from Free to Pro 500, with the allowance estimates reported at DevDay and the overflow credit value.
- CSV exports. Five CSV files for scripts or Google Sheets, with a note on importing without breaking formulas.
- Invoice and review guides. A one-page guide to reading a provider invoice and a 15-minute monthly review checklist.
How to Use
- Open the xlsx in Excel, Numbers or Google Sheets and read the Start Here sheet.
- On Prices, check the models you use against each vendor page and edit the yellow cells.
- On Agent Budget, replace the six example agents with your own and set a monthly cap per agent.
- Open Cache Sensitivity with your top agent's token counts to see what a higher hit rate saves.
- Run Break Even for any agent you plan to keep.
- Set provider billing alerts at the thresholds on the Alerts sheet and review monthly.
Who This Is For
- Developers running coding agents who want a number before the invoice arrives.
- Founders and small teams putting agents into support, research or review workflows.
- Finance and operations leads who need a defensible monthly AI budget.
- Consultants who quote agent projects and must show running costs to clients.
Prices change fast, so every row carries a date and you can edit it in one cell. Browse more in the AI agent tools collection, or try Agent Run Cost Simulator.
Ai-Agents-Workflows