Track Total AI Spending: Claude, OpenAI, Cursor
Short answer: to track your total AI spending, add up three things once a month: fixed subscriptions, the overage charged on top of them (Claude usage credits, Copilot additional usage, Cursor on-demand usage), and pay-per-token API bills. Each provider shows only its own share, so a spreadsheet does the totalling. For Claude Code and Codex, FavTray shows a running cost estimate in your Mac’s menu bar.
TL;DR
- Every provider has its own usage page. The table below lists where each one is, checked on 30 September 2026.
- Overage is the line that surprises you. Subscriptions are fixed; usage credits, AI Credits and on-demand usage are not.
- The one sourced benchmark: Anthropic says Claude Code averages about $13 per developer per active day, $150-250 per month, across enterprise deployments (Anthropic).
- FavTray estimates Claude Code and Codex cost from local logs and shows plan usage for other tools. Tracking is free on every plan. It has no dollar budget alerts.
- Run a 15-minute audit on the same day each month, and set a cap on every variable line.
Why is it hard to see your total AI spend?
Because you pay in three different ways, and each provider shows only its own part. Subscriptions are a fixed monthly fee. Overage on top of a subscription is metered in the provider’s own unit: Claude usage credits, GitHub AI Credits, Cursor on-demand usage. API access is billed per token. None of the dashboards know about the others, so there’s no total unless you add it up yourself.
The units don’t line up either. Claude Code reports tokens and an estimated dollar figure, Copilot reports AI Credits, Cursor reports usage against an included amount, and the API consoles report dollars. The only common unit is what lands on your card, which is why the monthly audit below works from dollars.
Where do you check usage and spend in each AI tool?
Each tool keeps usage in a different place, and most of them also let you cap overage. Plan prices change often, so they’re in our AI coding assistant cost comparison rather than here.
| Tool | What you pay for | Where to see usage and spend | Where to cap it |
|---|---|---|---|
| Claude API | Per token | Claude Console Usage and Cost pages. console.anthropic.com now redirects to platform.claude.com | Console Settings > Billing > Spend limits (docs) |
| Claude Pro / Max | Subscription, plus usage credits if you turn them on | claude.ai Settings > Usage: plan limits, credit balance and this month’s usage-credit spend | Monthly spend limit in the same Usage credits section |
| Claude Code | Your Claude plan or your Console API key | /usage in the CLI: a session cost estimate (API users), plan usage bars (subscribers) and a usage-credits row (docs) | claude -p --max-budget-usd 5.00 "..." stops a scripted run at a dollar cap |
| GitHub Copilot | Subscription with monthly AI Credits, since 1 June 2026 (GitHub) | github.com/settings/billing > AI usage, or profile picture > Copilot settings > Usage (docs) | A budget for additional usage in billing settings |
| Cursor | Subscription with included usage | cursor.com/dashboard/usage; on-demand charges under Billing & Invoices (docs) | Dashboard > Spending tab > monthly spend limit (docs) |
| ChatGPT and Codex | Subscription, plus credits on Plus and Pro | ChatGPT Settings > Billing for the plan; Codex limits at chatgpt.com/codex/settings/usage or /status in the Codex CLI (docs) | Credits are bought separately from the plan |
| OpenAI API | Per token | platform.openai.com/usage. The Costs tab reconciles to your invoice | Spend alerts and a hard spend limit on the Limits page (docs) |
| Gemini API | Per token | Google AI Studio Dashboard > Usage; dollars in the Cloud Billing console | Monthly spend cap per project on the AI Studio Spend page (docs) |
Two details that catch people out. Copilot code completions and next edit suggestions don’t use AI Credits; chat, agent mode, the CLI and the cloud agent do (GitHub docs). Annual Copilot subscribers stay on the older premium-request pricing until their plan ends. And spend limits aren’t instant: Cursor warns that usage “can briefly exceed your spend limit” before it stops.
How does FavTray track Claude Code and Codex costs on a Mac?
FavTray’s AI Usage tool estimates what Claude Code and Codex cost by reading their local logs on your Mac (~/.claude/projects and ~/.codex/sessions) and pricing the tokens at list rates. The menu bar popover shows today’s cost and the last 30 days; the provider screen adds 7-day, average-day and peak-day figures and a day-by-day history. For the other providers it supports, including Cursor, Copilot, Gemini and Windsurf, it shows how much of your plan’s usage is left and when it resets.
- API spend with an Admin key. Add an Anthropic or OpenAI Admin key and FavTray shows that organisation’s API spend for today and the last 30 days, read from each provider’s cost API. Anthropic’s Admin API isn’t available to individual accounts (Anthropic). The keys stay in FavTray’s own Keychain item.
- Free on every plan. Usage and cost tracking, with full history, needs no Lifetime licence.
- No budget alerts. FavTray doesn’t alert you at a dollar amount. Use the provider caps in the table above for a hard stop.
- Estimates, not invoices. If Claude Code runs on a Pro or Max plan, the dollar figure is what those tokens would cost at API rates. You’re charged for the plan and any usage credits, not that figure. Anthropic says the same about the session cost in
/usage.
See the AI Usage Tracker for what it shows, or the Claude cost tracking guide for the Claude side in depth.
What does a 15-minute monthly AI cost audit look like?
Pick one day each month and work through six steps: fixed costs, overage, API bills, local estimates, the total, then a decision. Doing it on the same day makes month-to-month changes easy to spot.
1. List fixed subscriptions (3 min). Search your card statement and any app-store subscriptions for Anthropic, OpenAI, GitHub, Cursor and Google. Write down each plan and what it cost this month.
2. Check overage and usage-based charges (4 min). These are the lines that move:
- Claude usage credits: claude.ai Settings > Usage shows this month’s usage-credit spend.
- Copilot additional usage: github.com/settings/billing > AI usage shows included credits used and any additional usage. On an annual plan, check premium requests instead.
- Cursor on-demand usage: the dashboard’s Billing & Invoices tab.
- ChatGPT or Codex credits you bought this month.
3. Pull API bills (3 min). Claude Console Cost page, the Costs tab on platform.openai.com/usage, and Cloud Billing for Gemini.
4. Check the local estimates, but don’t add them (2 min). Read FavTray’s 30-day Claude Code and Codex totals, or /usage in Claude Code. If Claude Code signs in with a Console API key, that spend is already in step 3. On a subscription, the estimate isn’t a charge. Use it to see which tool and which days drove your usage.
5. Fill in the table (1 min).
| Category | This month | Last month | 3-month average |
|---|---|---|---|
| Fixed subscriptions | $____ | $____ | $____ |
| Overage (usage credits, AI Credits, on-demand) | $____ | $____ | $____ |
| API bills | $____ | $____ | $____ |
| Total | $____ | $____ | $____ |
6. Decide (2 min).
- Cancel any subscription you haven’t opened in two weeks.
- Overage every month? Compare the next tier up; for Claude, see Pro vs Max vs API.
- API spend climbing? Work through ways to cut Claude Code token usage.
- Any variable line without a cap? Set one now.
How do you set an AI tool budget?
Start from your own numbers, not an industry average. After two or three audits you’ll have a baseline from the table above. Set your budget there, then put a cap on each variable line so no single tool can blow through it.
The one published benchmark worth knowing comes from Anthropic: across enterprise deployments, Claude Code averages about $13 per developer per active day and $150-250 per developer per month, and 90% of users stay under $30 per active day (Anthropic). That’s spend at API rates for work-day use. On a subscription you pay the plan price plus any usage credits instead.
- Cap the variable lines. Every overage and API line in the where-to-check table has a cap. Set each one to the most you’d pay without a second thought.
- Keep fixed costs to tools you use weekly. A subscription you open once a month costs the same as one you use every day.
- Review each quarter. Raise a cap only when you hit it and the work was worth it.
What tools help consolidate AI spending in one place?
No tool covers subscriptions, overage and API bills together. The practical setup is live tracking for the variable costs plus a spreadsheet for the fixed ones. For a comparison of tracking apps, see best AI usage tracking tools.
| Spending | Where to track it | Effort |
|---|---|---|
| Claude Code and Codex usage | FavTray (live estimate), or /usage in Claude Code and /status in Codex | Automatic |
| Plan usage in Cursor, Copilot, Gemini, Windsurf and others | FavTray, or each tool’s own usage page | Automatic |
| Claude API and OpenAI API spend | Provider dashboards, or FavTray with an Admin key | Automatic |
| Fixed subscriptions | A spreadsheet or budgeting app | 3 min a month |
If you’d rather build your own dashboard, both API providers publish daily spend to Admin keys. Anthropic’s Cost API returns amounts in cents, as decimal strings:
curl "https://api.anthropic.com/v1/organizations/cost_report?starting_at=2026-09-01T00:00:00Z&ending_at=2026-10-01T00:00:00Z&bucket_width=1d&limit=31" \
-H "anthropic-version: 2023-06-01" \
-H "x-api-key: $ANTHROPIC_ADMIN_KEY"OpenAI’s Costs endpoint takes a Unix start time and returns dollar amounts:
curl "https://api.openai.com/v1/organization/costs?start_time=$(( $(date +%s) - 30*86400 ))&bucket_width=1d&limit=30" \
-H "Authorization: Bearer $OPENAI_ADMIN_KEY"Subscriptions have no equivalent API, so the spreadsheet stays either way.
Frequently Asked Questions
How do I find out how much I spend on AI tools each month?
Add up three things: your fixed subscriptions (from your card statement), any overage on top of them (Claude usage credits, Copilot additional usage, Cursor on-demand usage, ChatGPT credits), and pay-per-token API bills from each provider's usage page. Each provider shows only its own share, so a spreadsheet does the totalling. The only published benchmark with a source is Anthropic's: across enterprise deployments, Claude Code averages about $13 per developer per active day and $150-250 per month.
Is there a single dashboard to track all AI tool costs?
No dashboard pulls subscription billing from Claude, OpenAI, Cursor and Copilot into one view. FavTray gets closest on a Mac: it estimates Claude Code and Codex costs from their local logs, shows how much plan usage is left in Cursor, Copilot, Gemini, Windsurf and other tools, and with an Anthropic or OpenAI Admin key shows your organisation's API spend. Fixed subscription fees still go in a spreadsheet.
What counts as overage on Claude, Copilot and Cursor?
On Claude Pro and Max it's usage credits, billed at API rates once you pass your plan's limit and shown in claude.ai under Settings > Usage. On GitHub Copilot it's additional usage beyond your plan's monthly AI Credits (1 credit = $0.01), which you allow by setting a budget. On Cursor it's on-demand usage, billed at API rates after your included usage runs out, with a monthly spend limit on the dashboard's Spending tab.
Does FavTray alert me when I pass a dollar budget?
No. FavTray shows running cost estimates and plan usage, but it has no dollar budget alerts. For a hard stop, set the caps each provider offers: a spend limit in the Claude Console, a monthly limit on Claude usage credits, a Copilot budget, a Cursor spend limit, or a hard spend limit on the OpenAI platform.
Should I consolidate onto fewer AI tools to save money?
Only if you're paying for overlap. Each tool is good at something different: Copilot and Cursor at in-editor completions and edits, Claude Code and Codex at longer agentic tasks. Check which subscriptions you actually opened in the last two weeks and cancel the ones you didn't, rather than dropping a tool you use daily.