Field note
Claude Usage Credits: Budget the Overflow
Usage credits let a paid Claude plan keep working after the included allowance runs out, so the real decision is who may spend them and how much.
Usage credits let a paid Claude plan keep working after the included allowance runs out, so the real decision is who may spend them and how much.
How usage credits work
A Pro or Max subscription includes an allowance that is metered in a rolling five-hour window and a weekly window. When a person hits the included limit, Claude normally stops until the window resets. Usage credits, which Anthropic's older material calls extra usage, remove that stop. At the time of writing they are available on Pro, Max 5x and Max 20x, and on Team and Enterprise plans through the admin console.
Four facts shape every budget decision:
- Credits are billed at standard API rates, per token. The subscription price does not change, but the overflow is pay-as-you-go.
- Credits are only drawn once the included allowance in the current five-hour window is used. They do not change the reset timing.
- You fund them by adding a prepaid amount or turning on auto-reload, and you can set a monthly spending cap. Anthropic also states a daily redemption limit of 2,000 US dollars at the time of writing.
- In Claude Code, the prompt cache lifetime drops from one hour to five minutes once a session is drawing on credits, so a long idle gap costs more than it did within the allowance.
Credits do not expire in the UK at the time of writing, but check the support article before relying on that.
Where to see the spend
Individuals manage credits at Settings, Usage on claude.ai. That page shows the credit balance, this month's spend and the monthly cap. Mobile subscribers need the web version to enable credits.
In Claude Code, /usage shows a usage-credits row with the month's spend against the cap. Running /usage-credits opens the same settings page for Pro and Max, opens Admin settings, Usage for Team and Enterprise members with billing access, and for members without billing access sends a request to the organisation's admins after a confirmation step.
On Team and Enterprise, admins can set spend limits at organisation, group or individual level, and the spend report in organisation analytics shows estimated spend per user and per model with a CSV export, updated daily. Usage inside the seat allowance is not metered in pounds or dollars; only the credit overflow is.
Anthropic's published figure for enterprise Claude Code deployments is around 13 US dollars per developer per active day, with 90 percent of users under 30 dollars. Treat that as a planning anchor, not a promise. A coding seat costs more than a chat seat because each turn carries file contents and tool output.
A planning table for a small team
Start from who is likely to hit the allowance and what it costs the business when they stop. Then set caps that match.
| Role | Typical pattern | Suggested credit policy |
|---|---|---|
| Owner or director using chat | Short questions, occasional documents | Credits off; wait for the reset |
| Marketing or operations, daily chat and Projects | Long documents, repeated drafts | Credits on with a low monthly cap, for example one day of API spend |
| One developer on Claude Code | Long sessions, large repositories | Credits on, cap at two to four times the seat price, review monthly |
| Two or more developers on Claude Code | Parallel sessions, automation, subagents | Team plan, org-wide cap plus per-member caps, spend report reviewed weekly |
| Automation or scheduled jobs | Unattended, repeatable | Do not run on a personal plan; use the API with a workspace spend limit |
The last row matters. Unattended jobs belong on the Claude Console with their own key and workspace limit, so they cannot silently drain a colleague's credits.
What to do when someone hits the limit
- Read the message. "You've hit your session limit" or "weekly limit" is the allowance. "You've hit your individual spend limit" or "org's monthly spend limit" means credits are on and a cap has been reached.
- Check the timer. The message shows when the window resets. If the reset is close, the cheapest answer is to wait. On Claude Code v2.1.234 or later,
/rate-limit-optionscan wait and continue the interrupted task automatically. - If the work is genuinely urgent, an admin raises the cap the message names in Admin settings, Usage. Record why.
- Fix the cause before the next month. In Claude Code,
/usageflags behaviour such as long context or cache misses when it accounts for 10 percent or more of recent usage./clearbetween unrelated tasks and a smaller model for routine work reduce the overflow more than any cap does.
Our note on the five-hour and weekly limits covers the allowance side in more detail. For the wider question of what happens to a workflow when an AI tool stops, see AI chatbot usage limits and business continuity.
Controls worth writing down
- One named person approves turning credits on, and the approval says the cap.
- Caps are per person on Team, not only org-wide, and the spend report is reviewed on a fixed day each month.
- Anyone with an
ANTHROPIC_API_KEYset in their shell is billed to the API, not the subscription./statusin Claude Code shows which authentication is active. Check this before blaming credits for an unexpected bill. - If a member leaves, their credit caps and requests are removed with the seat.