Codex Subscriptions: Limits, Credits, and the Cost of Upgrading
Subscription guides
A limit warning makes an upgrade look like the obvious next step. But the problem might be a short usage window, a periodic allowance, or a separate API bill. Paying the wrong account will leave the original problem intact. Start with the active session, then compare the costs. Plan facts were checked on October 9, 2026; the work and income figures below are hypothetical.
Identify the account paying for this session
A subscription is an account entitlement; an API key is a separate authentication and billing route. Inspect the running session rather than the membership page you remember opening. In the Codex CLI, start with /status and compare it with the usage page for the same account. In the desktop app, inspect the signed-in account, workspace, and usage window.
If you use both a personal account and a company workspace, establish which workspace owns the task. Renewing a personal plan cannot replenish a company allowance. Funding an API account does not reset a ChatGPT subscription allowance either. The table locates the purchase route; it does not imply transferable balances.
| Route | What you pay for | Where to check |
|---|---|---|
| ChatGPT sign-in | That account or workspace’s plan access | Account, workspace, plan, usage |
| Extra credits after included usage | Eligible additional usage | Same account’s credit balance and terms |
| API key | Calls billed to the key’s API account | API organization / project usage and billing |
Message counts are a poor unit for a purchase decision
Official usage guidance accounts for model, context, and task complexity. Repeated file reads and tool output in a long session are different from a fresh session changing one sentence. Both may look like one request. API token prices also cannot tell you a fixed number of tasks included in a subscription.
Keep a representative week of limit events: model, speed mode, task type, used percentage, limit name, and reset time. Exclude client data and credentials. Five interruptions inside a short window suggest a different bottleneck from exhausting a weekly allowance early every week. On the review date, OpenAI listed a five-hour window for Plus and possible weekly limits. Pro currently has no five-hour limit; that does not imply unlimited access across every model and period.
The log can reveal avoidable work. A navigation-copy edit rarely needs the agent to re-analyse the whole project. A cross-directory refactor does need dependency and test context; stripping it can cause rework. Measure consumption per accepted change, rather than congratulating yourself on fewer requests.
Compare the bottleneck before comparing Pro tiers
On the review date, official monthly list prices were $20 for Plus and $100, $200, or $500 for Pro. These are USD list prices, distinct from shop quotes, local taxes, payment-channel costs, and upgrade settlement. An older comparison that describes Pro only as a $200 plan misses the current options.
Additional resources must address the limit you actually hit. Before paying, check that tier’s current allowances, supported models and speeds, and immediate-upgrade billing. Labels such as “5x” or “10x” do not establish proportional completed-project counts. Task duration and success rates do not scale with a plan name.
| Observation | First step | Question before purchase |
|---|---|---|
| Occasional burst | Check reset and one-off extra usage | Can this task wait for a reset? |
| Repeated interruptions across weeks | Match the limit to Pro allowances | Does the extra capacity address this bottleneck? |
| API-key sessions | Inspect API usage and budget | Will you switch authentication after subscribing? |
| Capacity remains; output fails | Improve scope and acceptance criteria | Is capacity actually causing the failure? |
Work an upgrade calculation without pricing every wait as income
Using list prices for illustration, Plus at $20 to Pro at $100 adds $80 for a full month. Suppose six urgent interruptions each cost 20 minutes: two hours total. If half of that time can be spent on other work, only one hour is lost. Valuing that hour at $100 would, in theory, cover the difference.
Three assumptions matter: the limit causes the wait, the new plan removes it, and the freed time has a useful destination. If the upgrade removes only half the interruptions, the benefit drops to $50, below the $80 difference. A larger subscription does not create the assumed income.
Estimate the loss you cannot work around, then compare the upgrade’s actual incremental cost with expected extra-credit spending. Price one-off capacity for a temporary sprint; compare recurring tiers for repeated excess. Use one currency and billing period, including taxes and upgrade settlement. Do not subtract a shop quote in CNY from an official USD price.
Avoidable loss = interruptions × time waiting
× fraction not usable elsewhere × value of time
Consider upgrading when:
avoidable loss > actual incremental upgrade cost
and the new tier addresses the observed limitVerify delivery and the session separately
My shop offers monthly ChatGPT Plus / Pro subscriptions for your own account; the product page lists available tiers, prices, and requirements. That product buys membership, with separate routes for extra Codex credits and API balance. The image below is a public product-page demo, not a customer order or proof of delivery.
Verification has two parts: check that the correct account has the new plan, then confirm Codex is using that account and workspace. An API-key session can still use API billing. Any change to periodic allowances should be checked in the resulting usage page rather than inferred from a successful payment.
If access still looks wrong, keep the order reference, plan state, and error message for order-page support. Distinguishing an undelivered plan from a delivered plan used through the wrong authentication route makes the issue easier to resolve.
- Record your account, current plan, and the specific limit before buying.
- Follow the product demo for ordering and submitting required material.
- After delivery, verify the plan, then Codex authentication and usage.
Workflow demo, not a real order or proof of fulfillment.
Use this product for membership subscriptions. API balance and extra credits are separate.