Codex Costs: API Billing vs Subscription Usage

Check how you are billed

OpenAI distinguishes subscription sign-in from API-key authentication. The former uses ChatGPT product access; the latter is billed as API usage. CLI users can check the active method with codex login status. Do not sign out or replace a key merely to understand a bill. Official authentication documentation

If your organization supplies a model gateway, it or the gateway vendor may manage billing. Identify the destination and payer before looking for the relevant records. A successful sign-in confirms authentication. Check model and feature access separately.

Compare Codex model prices on the same basis

These are standard text API rates for GPT-5.6 Sol and GPT-5.5, in US dollars per million tokens. They are not subscription prices or credit rates.

Sources: GPT-5.6 Sol model page and GPT-5.5 model page. The 5.6 Sol page says promotional pricing lasts at least through November 21, 2026. Check rates again for later use.

The table supports a direct comparison when the requests use the same amounts of each token category. Models may produce different output lengths or need different numbers of attempts. A lower token rate does not by itself establish a lower cost for the completed task.

A small calculation makes the token bill clearer

Suppose one short request uses 10,000 uncached input tokens and 2,000 output tokens, with no cache writes, tool charges, or other additions.

• 5.6 Sol: 10,000 ÷ 1,000,000 × 4 + 2,000 ÷ 1,000,000 × 20 = $0.08.
• 5.5: 10,000 ÷ 1,000,000 × 5 + 2,000 ÷ 1,000,000 × 30 = $0.11.

The calculation covers token costs under those assumptions. The number of messages visible in a chat is not enough to calculate a daily bill. One user task can contain multiple requests, tool results, and retries. Count the actual input and output of each request.

Sending the same material again does not prove that it was served from cache. Check the API usage fields or billing records. If cached tokens are included in total input, separate them before calculating to avoid billing the same portion at both rates in your estimate.

Why actual spend can be more complicated

The GPT-5.6 Sol page also specifies long-input and cache-write conditions. Requests exceeding 272K input tokens apply higher input and output rates to the whole request, and cache writes have separate pricing conditions. Those conditions require a different calculation from the short-request example above. Model-page pricing conditions

Check tool and other service charges separately. Local estimates can help manage a budget, but the relevant service's metering and billing records determine the bill. API list rates, gateway rates, and organizational contracts may differ.

Subscription limits are not a fixed number of calls

OpenAI's product documentation separates credits from API pricing and notes that credit prices do not themselves determine included subscription usage. Use the product usage dashboard to check limits; in a current Codex CLI session, follow the documented /status route. Official pricing and usage documentation

“This plan gives me a fixed number of complex tasks” is not a reliable budget. Long tasks, images, tool output, and reasoning settings change usage between tasks. Check the limits for your account, workspace, and product separately.

To plan subscription needs, track your task types, actual usage, and occasions when you hit a limit. Those records give you a better basis for estimating your workload than someone else's usage screenshot.

Record the cost of completing one task

For example, repair a function with an existing test. Record the starting code version and acceptance checks. Afterwards, record which models ran, how many requests were made, whether tests passed, and how much additional checking or editing you did.

Keep retries in the record. A lower-priced model can cost more overall if it needs extra attempts or extensive manual correction. Conversely, a task that is easy to verify does not automatically require the most expensive configuration. Choose by task rather than a permanent ranking.

DataCamp's Codex data-workflow tutorial offers useful task-organization context. Its Python setup and older examples are useful context, not evidence of current account limits or your actual bill. Existing online tutorial

Common questions

Does a ChatGPT subscription cover API charges?

Do not assume so. API-key use is metered through the corresponding API account; subscription sign-in follows product-access and usage rules. Check the active method first.

Does cheaper 5.6 Sol API pricing guarantee more tasks on my plan?

No. API rates do not directly establish included subscription capacity. Check product rules and the limits currently shown for your account.

Can I treat one credit as one dollar?

No. They are different units, with potentially different purchase conditions and applicable rates. Changing the currency symbol on a credits table does not make it an API price table.

Understand the bill before switching models

To understand Codex costs, identify who records your usage and which unit they use. Then compare models and tasks on the same basis. For GPT-6 selection, read the model-family guide. For API-key setup in DSH, use the OpenAI configuration guide.

Model setup, selection and cost guides