Claude Code, OpenAI Codex and Cursor each show usage somewhere, but the numbers answer different questions. A session cost, a plan-limit reading, a request count, a token total and an API invoice are not interchangeable. None of the vendor pages cited below reports developer time spent coding. If you want to know how long an agent ran, you need a clock. If you want to know what you were billed, you need the billing surface for your account type. Most apparent disagreements between these numbers come from mixing those up.
This guide treats the three agents as Claude Code, OpenAI Codex and Cursor, based on the vendor documentation available at the time of writing. Plan prices and quotas change, so check the live pages before repeating any figure.
As an Amazon Associate I earn from qualifying purchases.
Start with your billing mode
Every reading depends on how you pay. Anthropic’s help article on models, usage and limits in Claude Code says metering depends on how you sign in. Organization plan usage is an included pool that resets on a rolling basis. API-key use is billed per token to the Console or cloud account you configured.
Free tools Windows power users keep installed
One-click scans. No signup required.
Codex and Cursor follow the same logic: a plan allowance and API billing are separate ledgers. OpenAI’s plan guidance treats limits and credits apart from API activity, and Cursor meters agent usage against model and API pricing within its plans.
#1 Best Overall
Where each product shows usage
| Product | Surface | What the figure shows | Scope | Available to |
|---|---|---|---|---|
| Claude Code (plan sign-in) | Plan usage, as described in the Anthropic help article | Usage against an included pool with a rolling reset | Organization plan | Plan users; the cited page describes the metering rule rather than a specific screen |
| Claude Code | /cost in a session |
Token and dollar usage for the current session | Current session, API billing | API-key users |
| Claude Console | Cost and Usage Reporting | Charts with token and tool-use costs | Account | Specified Developer, Billing and Admin roles |
| Codex (ChatGPT plan) | Settings or the usage dashboard; /status in the Codex CLI |
Position against normal Codex limits, plus any credits or billing | Plan limits | Plan users; enterprise token-based billing controls are workspace-admin territory |
| OpenAI API | Usage Dashboard | Current and past billing periods, with project filters and exports | Organization; not combined across organizations | Account holders with API access |
| Cursor | Individual dashboard | Requests and token breakdowns | Individual account | Individual users |
| Cursor | Admin API and team analytics | Daily metrics and filtered usage events; team request categories and subscription or API usage fields | Team | Team admins |
Claude Code
The /cost figure is a session view for API billing
Claude Code’s /cost command shows the current session’s token and dollar usage, and Anthropic’s help article documents it in the API billing context. It is a useful check on one session. It is not a statement of a plan subscription charge, so a subscriber on an included plan should not read it as what they were billed.
Account-level reporting lives in the Console
For API-key use, Anthropic’s Cost and Usage Reporting in the Claude Console article describes charts with token and tool-use costs. Access is limited to specified Developer, Billing and Admin roles, so an individual developer may see a narrower view than the account owner.
Codex
Plan limits: Settings, the usage dashboard and /status
OpenAI’s guidance for using Codex with your ChatGPT plan tells users nearing or at a limit to check Settings or the usage dashboard. Inside an active Codex CLI session, run /status. Usage counts against normal Codex limits and any applicable credits or billing.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →API activity and cost
OpenAI’s reviewing API usage and costs article describes the Usage Dashboard for current and past billing periods, with project filters and exports. The dashboard does not combine usage or cost across organizations. It is the place to check API activity. The cited pages do not describe it as a complete record of plan-based Codex sessions.
Rank #3
Cursor
Individual dashboard: requests and tokens
Cursor’s Models and Pricing page says agent usage is metered according to model and API pricing, and that the model you select changes how included usage is consumed. The dashboard shows requests and token breakdowns. The plan quantities and examples on that page are specific to each plan and change over time, so confirm them before quoting.
Team reporting: Admin API and analytics
Team admins get more. Cursor’s Admin API documentation describes daily metrics and filtered usage events. For token-based events it identifies the model, usage category, request costs and token usage. Team analytics show request categories and subscription or API usage fields. These are admin capabilities, so an individual member should not assume they can see or export the same events. The team dashboard covers billing, usage and team settings, and billing documentation sits alongside it.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.The three ways to get your hours wrong
1. Reading a session or plan figure as a bill
A Claude Code session dollar amount, a Codex limit position and a Cursor request count each answer a narrower question than “what did I pay?” Before you compare anything, identify whether the number belongs to a plan allowance, a session, or an API or team billing account. Then check the invoice for that account, not the nearest number on screen.
2. Reading requests or tokens as time
A request is not an hour, and neither is a token total or a quota percentage. The vendor pages describe tokens, requests, limits, session cost and spend, but none gives a method for converting them into time. If you need hours, record elapsed time from your own clock or calendar, and label it as elapsed time rather than agent usage.
3. Comparing rows that do not share a scope
Two figures can look contradictory because they cover different models, windows or accounts. Claude Code plan usage uses a rolling reset, an API dashboard uses billing periods, and a team report covers members who may not be visible to you individually. A difference in any of those fields makes the rows non-comparable.
Quick Recap
A comparison method you can repeat
- Record the account type for each product: plan seat, API key, individual account or team.
- Record the surface you read: session view, plan-limit view, API dashboard, Console report, Cursor dashboard or admin export.
- Record the unit: tokens, requests, utilization percentage or currency.
- Record the window: current session, rolling reset, billing period or your chosen date range, plus the timezone you read it in.
- Record the model, since Cursor documents that model choice changes consumption.
- Mark any pair of rows that differs on a field above, or where a field is not visible, as non-comparable. Do not convert one unit into another.
What the vendor sources do not establish
- The three products are not documented as sharing a log schema, so do not assume field names, record formats or file locations match across them.
- Local session figures are not documented as reconciling exactly to billing. Use the billing surface for your account for invoice questions.
- No published rate of misreading, or typical size of discrepancy among these three products, was found in the vendor pages cited here. Treat any percentage you see elsewhere as unverified.
- Plan prices, included quantities and reset rules change. Check each vendor’s live page before relying on a figure.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
