What an AI cost audit looks like.
This is the shape of the report you receive. The team is invented: twelve people, eight of them engineers. Every price is real and is the one in the catalogue today.
Sample. The company and its usage are made up. Your report is written by hand for your own stack and usage.
6 things to look at. Plan prices in US dollars before tax, at the vendors’ published prices, plus API use at today’s rate.
1. What the team pays now
| Tool and plan | People | Per month | Price read |
|---|---|---|---|
| CursorTeams (Premium seat) | 8 | $240.00 | checked 2026-11-15 |
| GitHub CopilotBusiness | 8 | $152.00 | checked 2026-11-15 |
| ChatGPTBusiness | 12 | $300.00 | checked 2026-11-15 |
| ClaudeTeam (Standard seat) | 4 | $100.00 | checked 2026-11-15 |
| Claude Opus 5.5 API500 coding tasks a month | — | $2,670.00 | Vendor page · Nov 14, 2026 |
2. What we found
- 01
Two tools for one job
You pay for 2 coding assistants (Cursor, GitHub Copilot): $392.00 a month together. Teams often need more than one, but it is the first thing worth checking. In an audit we ask who uses which, and price the team on each one alone.
- 02
Two tools for one job
You pay for 2 general assistants (ChatGPT, Claude): $400.00 a month together. Teams often need more than one, but it is the first thing worth checking. In an audit we ask who uses which, and price the team on each one alone.
- 03
Monthly billing on plans you will keep
Paying yearly instead of monthly would save $960.00 a year on ChatGPT, Claude, at the vendors’ published annual prices. You commit for a year.
- 04
Costs the price list does not show
For Cursor, ChatGPT, Claude the vendor does not publish the usage allowance as a number, so the cost of heavy use is unknown: it can mean extra charges or a higher plan. We ask for your usage export and check it against what the vendor does publish.
- 05
API spend
The team’s agent runs on Claude Opus 5.5: $5,340.00 per 1,000 coding tasks at today’s price, about $2,670 a month at 500 tasks. The cheapest model with a verified price for the same task size is Ling 3.0 Tiny at $15.60 per 1,000. Cheapest is not equal: the audit prices the two or three models that fit the work and says what would have to be tested before a switch.
- 06
Dates to plan for
8 retirements or price changes are announced after today. The audit lists the ones your stack depends on, with the vendor’s suggested replacement.
3. What is about to change
The next announced dates in the catalogue. A real report keeps only the ones your stack uses.
| Date | Model | What happens | Suggested replacement | Source |
|---|---|---|---|---|
| Nov 30, 2026 | Claude Sonnet 4.5Anthropic | Retires | Claude Sonnet 5.5 | Vendor pagechecked Nov 14, 2026 |
| Nov 30, 2026 | GPT-5.6 SolOpenAI | Promotion endsThe $4 / $20 per 1M promotional price runs until 30 November. OpenAI does not publish the price that applies afterwards. | None named | Vendor pagechecked Nov 14, 2026 |
| Dec 1, 2026 | Gemini 4 ArgonGoogle | Price rises$2.00 / $10.00 → $4.00 / $20.00 per 1M tokens (input / output). | None named | Vendor pagechecked Nov 14, 2026 |
| Dec 1, 2026 | Gemini 4 CarbonGoogle | Price rises$2.00 / $10.00 → $4.00 / $20.00 per 1M tokens (input / output). | None named | Vendor pagechecked Nov 14, 2026 |
What a real report adds
- Your own usage, from the exports your tools provide, instead of list prices alone.
- Each change priced: what it saves a year, what it costs to make, and what should be tested first.
- Sources and check dates for every figure, so your finance team can verify it.