Guide

The Overage Switch: Which AI Subscriptions Keep Charging After You Hit the Limit

By Istar editors
At a glance

Compare Copilot, Claude, Cursor, and ChatGPT overage defaults, extra-spend behavior, and the cap or kill switch to check before included usage runs out.

Hitting an included usage limit is not one event. It is a fork. One vendor keeps running the meter. Another stops. A third waits for you to buy credits. The difference is not the marketing word “credits.” It is the default of the overage switch — and whether anyone has to opt in before a fixed subscription becomes a variable cost.

This is not a model-price roundup. It is a default-behavior audit of four products engineering teams actually buy: GitHub Copilot for organizations and enterprises, Claude paid plans, Cursor, and ChatGPT. Primary pages were checked on 28 August 2026. Billing copy, included pools, and toggle names move. Treat every default below as confirm-current against the live docs and the setting in your own account.

Primary sources (accessed 28 August 2026):

The table: same collision, four forks

Product What runs out What the official page says happens next Extra $ without a further human click? Cap / kill switch on the page
GitHub Copilot Business / Enterprise Pooled monthly AI credits Additional usage is enabled by default. If allowed, usage continues at published per-credit rates and is charged to the org or enterprise. If not allowed, usage is blocked until the pool refreshes. Yes, unless an admin has already disabled paid usage or a budget is already exhausted. Disable AI credits paid usage in AI Controls. Set user, cost-center, organization, and/or enterprise budgets.
Claude Pro / Max Included session usage With usage credits enabled and prepaid funds available, the user can choose to continue at standard API rates. Credits are billed separately from the subscription. No before setup. Enabling credits, adding funds, and enabling optional auto-reload are deliberate actions. Settings > Usage: disable credits or set a monthly spend limit; leave auto-reload off for a prepaid-only ceiling.
Cursor Pro / Pro Plus / Ultra Included usage in two monthly pools After included usage, you can add on-demand usage at the same API rates or upgrade. The page does not state whether on-demand is on by default. Not stated. Do not assume either default. Usage dashboard / editor settings; add or withhold on-demand. Start (India) does not include on-demand.
ChatGPT personal Included plan limits for supported features Credits extend usage after included limits. Hitting an eligible limit can show an option to add credits. Automatic reload is something you turn on if available. Not by default, unless automatic reload is already on. Buying credits is an explicit action. Settings > Usage or Usage & Billing. Leave automatic reload off if you want a hard stop after prepaid credits.
ChatGPT Business Per-seat included limits Credits are optional. If a user exceeds the seat limit and the workspace has no credits, the feature is blocked. Owners can buy credits and turn on automatic recharge. No, unless the workspace already holds credits or auto-recharge is on. After credits exist, seat and user limits are unspecified by default. Workspace settings → Billing: add credits, configure automatic reload and its monthly recharge limit, then set seat and user credit limits.
ChatGPT Enterprise / Edu Shared credit pool When the pool is exhausted, advanced features are paused unless owners enable overages or purchase more credits through the OpenAI account team. Only if overages are enabled or more credits are purchased. Do not enable overages; buy or allocate credits deliberately; watch threshold alerts.

Read the table as a default audit, not as a ranking. A vendor that stops at the limit is cheaper only if stopping is acceptable. A vendor that continues is only dangerous if nobody owns the cap.

GitHub Copilot: the meter stays on unless you turn it off

For Copilot Business and Copilot Enterprise, GitHub measures billable AI usage in AI credits. Input, output, and cached tokens are priced by model and converted into credits at a fixed rate: 1 AI credit = $0.01 USD.

Each assigned license contributes a monthly amount of included credits, pooled at the billing entity rather than kept as private buckets. GitHub’s example has 100 Copilot Business users sharing 190,000 credits. Power users can draw more while light users offset them. Mid-cycle license additions grow the pool immediately; removals do not shrink it until the next cycle. Unused included credits do not carry over, and the pool resets at 00:00:00 UTC on the first day of each calendar month.

The page listed standard included amounts of 1,900 credits per user per month for Business and 3,900 for Enterprise. Existing customers receive a promotional 3,000 / 7,000 for the first three months of usage-based billing, from 1 June through 1 September 2026. After that period, included usage returns to the standard amounts. If you are reading after 1 September, do not use the promotional numbers.

Not everything is on the meter. Copilot features that call AI models consume credits, including Copilot Chat, Copilot CLI, Copilot cloud agent, Copilot Spaces, Spark, and third-party coding agents. Code completions and next edit suggestions are not billed in AI credits and remain unlimited on paid plans.

The overage switch is the part finance teams miss:

  • Additional usage allowed: usage continues at published per-credit rates; extra spend is charged to the organization or enterprise.
  • Additional usage not allowed: usage is blocked until the next billing cycle.

GitHub states that additional usage is enabled by default for organizations and enterprises. Preventing spend beyond included credits requires an administrator to explicitly disable the AI credits paid usage policy in the enterprise’s or organization’s AI Controls settings.

The page’s second control layer is budgets, set in US dollars while usage is shown in credits. A $10 budget covers 1,000 credits:

  • User-level budgets cap one person’s consumption from the shared pool and additional usage. A $0 user-level budget blocks that user immediately.
  • Cost-center budgets cap metered charges for a group after the pool is exhausted.
  • Enterprise spending limits cap total metered charges after the pool is exhausted.
  • Organization-level budgets cap metered charges for seats billed to that organization after the pool is exhausted.

If a user-level budget is exhausted, that user’s Copilot access is halted even if the org pool still has capacity. An enterprise spending limit can also block a user before the personal budget is reached. GitHub states there is no automatic fallback to lower-cost models when a budget is exhausted.

Close / cap path: AI Controls → disable AI credits paid usage if you want a hard stop at included credits; then set user and entity budgets so that paid usage cannot run open-ended. Confirm the live account state, not just this article.

Claude: credits require setup, funds, and a choice to continue

Anthropic’s help article loaded successfully on this review pass. It says usage credits apply to Pro, Max 5x, and Max 20x. When included limits are reached, a user with credits enabled and funds available can choose to continue; subsequent usage is billed at standard API rates, while the normal session limit still resets every five hours. Anthropic explicitly says these charges are separate from the paid Claude subscription.

The setup is deliberate: open Settings > Usage, locate Usage credits, click Enable, add a payment method, configure spending preferences, set a monthly spend limit, and prepay funds. Auto-reload is optional. The same page gives a $2,000 daily redemption limit, which is a platform ceiling rather than a sensible personal budget.

That makes the default auditable: a new subscription does not silently become usage-credit billing. Someone must enable the feature and fund it. After setup, however, optional auto-reload can turn a prepaid balance into a continuing purchase path, so the monthly limit still matters.

For the plan trade-off and a worked credit-cost model, see Claude Pro’s $20 Wall: Wait, Upgrade to Max, or Buy Usage Credits?.

Close / cap path: Settings > Usage → disable Usage credits for the included-plan hard stop, or set a monthly spend limit and leave auto-reload off. Mobile subscribers must enable and buy credits on the web app.

Cursor: on-demand is documented; the default is not

Cursor’s models-and-pricing page describes two monthly pools that reset with the billing cycle: Cursor Models and Other Models. Pro, Pro Plus, and Ultra include both pools, with the option to pay for additional usage. Both pools are visible in editor settings and on the usage dashboard.

The page listed Pro $20/month, Pro Plus $60/month, Ultra $200/month, and India-only Start ₹649/month including tax. Start covers the Cursor Models pool and does not include the Other Models pool, on-demand usage, Bugbot, Auto, Automations, or the Cursor SDK.

Under What happens when I reach my limit?, Cursor lists two options:

  1. Add on-demand usage — continue at the same API rates with pay-as-you-go billing.
  2. Upgrade your plan — move to a higher tier for more included usage.

The page says on-demand usage is billed monthly at the same rates and requests are never downgraded in quality or speed. It does not say whether on-demand is enabled by default when included usage hits zero. “Add on-demand usage” is an action, but that is neither an explicit hard stop nor GitHub’s explicit enabled-by-default policy. Confirm-current in the dashboard.

Teams and Enterprise third-party model requests also include a Cursor Token Rate of $0.25 per million tokens on top of model API pricing for included, on-demand, and BYOK usage. First-party Cursor models, including Grok and Composer, are exempt. That is an extra line item on business plans, not the overage switch itself.

Close / cap path: open the usage dashboard and editor settings. If on-demand is present, record whether it is on and whether a spend cap exists beside it. If you are on Start, the page says on-demand is unavailable; upgrading is the documented path to that option.

ChatGPT: credits are opt-in; auto-reload and overages are the switches

OpenAI’s personal-plan article says credits extend supported features after included limits are reached. If credit purchases are available, the account can show Add credits and expose the balance and purchase controls in Settings > Usage or Usage & Billing. Buying credits does not change the subscription plan or increase its included allowance. Availability, amounts, prices, payment methods, spending limits, and eligible features vary by account, region, and plan.

If automatic reload is available, OpenAI says you can turn it on from the same surfaces. That wording is opt-in. When enabled, a balance below the chosen minimum triggers a purchase back to the target balance, up to any maximum monthly spend set for automatic reload. One-time purchases sit outside that auto-reload maximum.

Workspace plans are a different machine.

On ChatGPT Business, users have per-seat limits for advanced features. If a user exceeds the limit and the workspace has purchased credits, they can continue from the shared pool. If no credits are available, the feature is blocked and users can ask an owner to add more. Owners buy credits at Settings > Billing and can configure automatic recharge.

The spend-control page adds an important second step. A workspace can set the minimum balance, target balance, and monthly recharge limit for auto-reload; leaving the recharge limit blank allows unlimited automatic reload purchases. OpenAI also says all seats and users have no credit usage limits specified by default. Owners may set limits by seat type and per-user override. The Business fork is therefore: no credits means blocked; credits or auto-reload without explicit limits can become uncapped variable usage.

On Enterprise and Edu, once the shared credit pool is exhausted, advanced features are paused unless owners enable overages or buy additional credits through the OpenAI account team. Usage limits control how the existing pool is used; they do not add credits.

Close / cap path:

  • Personal: Settings > Usage or Usage & Billing — keep automatic reload off unless you want a running purchase path.
  • Business: Workspace settings → Billing — set a non-blank monthly recharge limit and explicit seat or user credit limits.
  • Enterprise / Edu: do not enable overages unless finance has approved variable spend; buy credits through the account team if the shared pool needs to grow.

This article does not treat OpenAI API prepaid billing as the same product. API credit exhaustion and ChatGPT subscription overage are different meters.

How to read a default

Three questions separate a subscription from a surprise invoice:

  1. When included usage hits zero, does work continue, stop, or wait for setup? GitHub says extra usage is on by default for organizations and enterprises. Claude and ChatGPT personal require deliberate credit setup; ChatGPT Business blocks without a credit pool; Enterprise and Edu pause unless owners enable overages. Cursor documents an add-on without stating the default.
  2. Is extra usage on a separate bill or mixed into the seat invoice? GitHub charges additional spend to the organization or enterprise. Anthropic says usage credits are separate from the paid subscription. OpenAI credits are an add-on rather than more included plan usage. Cursor bills on-demand monthly at API rates.
  3. If extra usage is on, is there a cap that is on by default? GitHub offers budgets but paid usage starts on. Anthropic asks users to configure spending preferences and a monthly limit. ChatGPT Business seat and user limits start unspecified. Cursor’s page does not document a default on-demand cap. An optional cap is not protection until someone sets it.

Assume these pages can change the toggle name, included pool, or default. Re-open the source pages and in-product screens before a quarterly close, seat expansion, or coding-agent rollout. Agent features are how these meters jump from “chat” to unattended token burn.

Check these switches today

  1. GitHub Copilot: AI Controls → AI credits paid usage — is additional usage still enabled, and is there a dollar budget behind it?
  2. Claude: Settings > Usage → Usage credits — is it enabled, is auto-reload on, and is the monthly limit deliberate?
  3. Cursor: Usage dashboard → on-demand usage — is it on, and will the next request after included usage bill at API rates?
  4. ChatGPT: personal Settings > Usage / Usage & Billing, or workspace Billing — is any auto-buy path on without a monthly cap?