Codex usage can involve tokens, credits, time-window limits, a ChatGPT subscription, or API billing. These measures are related, but they are not interchangeable.
OpenAI explicitly warns against using API token prices to estimate included ChatGPT tasks.
Four separate measures
| Measure | Meaning | Practical effect |
|---|---|---|
| Tokens | Model input and output units | More context or output usually increases consumption |
| Credits | A pricing unit for supported usage | Model and token type affect the charge |
| Usage limit | Included activity during a time window | Tasks can pause until reset or continue with credits |
| Subscription | The recurring ChatGPT plan price | It stays fixed until the plan changes |
A token reduction does not automatically reduce a subscription price. It can make an allowance last longer or reduce metered spend.
Which Codex windows apply
OpenAI currently publishes five-hour usage estimates for Plus and Standard Business. Pro plans currently have no five-hour limit.
Weekly limits may also apply. The usage dashboard is the source of truth for your account’s current limits and reset times.
Published message counts are estimates. They change with the model, task size, context, reasoning, tools, retrieval, caching, and execution location.
Which products share usage
OpenAI says local Codex messages and cloud chats share the plan’s usage allowance. Cloud tasks can use more allowance than local messages.
ChatGPT Work and Codex also share usage, credits, and limits. An activity in one surface can reduce the allowance visible in another.
An API key uses separate API billing. It does not consume included ChatGPT tasks in the same way.
Why usage can rise quickly
Prompt length alone does not predict usage. These factors can matter more:
- the selected model;
- current conversation context;
- reasoning effort;
- local or cloud execution;
- tool calls and retrieved content;
- cached and uncached input;
- parallel workers and automations;
- output length.
A short follow-up can carry a large conversation behind it. A multi-step task can also make several model requests from one user instruction.
How to inspect current usage
Open the Codex usage dashboard for current account limits, credits, and reset times.
In the Codex CLI, /status shows current session information. Use the dashboard
for account-specific plan data because published estimates are not fixed limits.
What happens after included usage ends
Supported personal and business plans can extend usage with credits. Included usage is consumed first, then eligible work can draw from the credit balance.
The OpenAI API follows separate token pricing and rate limits. A ChatGPT plan does not convert API prices into a fixed number of included tasks.
Review the account’s credit balance and spend controls before starting long automations. This prevents an unexpected transition to metered usage.
How to make the allowance last longer
Use the lowest-cost configuration that still passes your checks:
- Start a new chat when the task changes.
- Keep repository instructions short and scoped.
- Select a smaller model for clear, routine work.
- Lower reasoning effort for simple tasks.
- Disable tools that the task does not need.
- Limit large tool results and repeated file reads.
- Stop unused workers and automations.
- Review usage after a representative task.
These steps can reduce consumption. They do not promise a fixed percentage or identical output from every configuration.
What our benchmark does and does not show
Our Codex SWE-rebench checkpoint compared complete workflows across 14 tasks. Each task ran three times.
Both arms passed 25 of 42 runs. Halv recorded $144.67 in selected model cost, compared with $337.50 for vanilla Codex.
The Halv workflow used 3.49 times as many tokens. Different model routing and caching produced lower recorded cost with more tokens.
The result supports a limited workflow cost claim. It does not establish fewer tokens, identical quality on every task, or universal savings.
Where Halv fits
Halv processes context on your device before supported coding agent requests leave the machine. It reports the token changes it measures.
Results vary with the request. Always compare the completed work, verifier result, total workflow cost, and total tokens.
Start a 7-day free trial. Halv Unlimited costs US$10 per month after the trial. A credit card is required. Cancel during the trial to pay nothing.
Sources
For a practical workflow, read how to reduce Codex usage. It covers task routing, account meters, and accepted work per allowance.
Official OpenAI documentation checked on October 14, 2026:
For a broader product comparison, read Claude Code vs Codex.