GPT Codex Explained: Access, Costs, and Your First Coding Task
“GPT Codex” usually means using OpenAI's Codex coding agent with GPT models and a ChatGPT account. It is not a separate subscription name. Codex can work with project files and development tools; the useful outcome is a change you can inspect and test, not merely a code snippet.
Contents
- GPT, ChatGPT, and Codex: what is the difference?
- How to access Codex now
- What does Codex cost?
- A small task you can reproduce
- Review results and control access
- Frequently asked questions
GPT, ChatGPT, and Codex: what is the difference?
| Term | Meaning in this guide | What to check |
|---|---|---|
| GPT model | The model doing reasoning and generating responses | Which model your account and interface offer |
| ChatGPT | The product and account used to access available modes | Your plan and selected mode |
| Codex | The agent workflow for software work | Project access, tools, permissions, and results |
| API key | A separate authentication and billing route | API costs and feature availability |
Do not treat “ChatGPT thinks, Codex executes” as a hard technical boundary. Codex can also analyze and plan. The practical distinction is whether the session has the relevant files and tools to complete your task.
How to access Codex now
The official quickstart describes selecting Codex from the ChatGPT dropdown for software development, opening a project or folder, and signing in with your ChatGPT account. It also covers CLI and IDE options. Availability depends on your plan and environment; an ordinary chat does not automatically have access to every local repository.
- Use the official installation or access instructions for your device.
- Sign in and select the intended project. Start with a disposable example or a version-controlled copy.
- Ask it to identify the relevant files and existing test command before editing.
- Confirm that it can actually access the project. Pasted code alone does not connect the rest of your repository.
For remote work, check the selected cloud environment and repository separately. Opening a cloud task does not automatically expose files from your computer.
What does Codex cost?
Checked November 13, 2026: official pricing lists Codex in ChatGPT plans, including Free and Go with narrower desktop access subject to rollout. Plus is listed at US$20/month; Pro starts at US$100/month with higher tiers. Paid access is not unlimited.
ChatGPT Work and Codex share usage. Task size and model choice affect consumption; use the account usage dashboard, or /status in the CLI, rather than assuming a fixed number of tasks. API-key usage has separate token billing and feature restrictions. Do not calculate included subscription tasks from API token prices.
For more detail, see our Codex usage-limit guide. Check your own account's offer and local checkout before upgrading.
A small task you can reproduce
Use this exercise in a disposable Python project. It is an example, not a claim that we tested a particular model's performance.
Create prices.py with:
def total_cents(prices):
return sum(prices)Then give Codex this task:
In prices.py, keep total_cents and its existing behavior for nonnegative integer prices. Reject booleans and all other types with TypeError; reject negative integers with ValueError. An empty list must return 0. Use no third-party dependencies. Add unittest coverage in test_prices.py. Run python -m unittest -v and report the actual result, changed files, and anything you could not verify. Do not change unrelated files or publish the project.
The acceptance checks are concrete: [100, 250] returns 350; [] returns 0; [-1] raises ValueError; [True], [1.5], and ["100"] raise TypeError. In Python, booleans are a subtype of integers, so a simple integer check can miss that requirement.

Read the changed files and test output. “Tests added” is different from “tests passed.” If Python or dependencies are unavailable, the result should say so. Run the checks in a working environment before accepting the change.
Review results and control access
Check the diff for unrelated changes, verify failure cases, and keep the previous version until you accept the result. For a real application, also exercise the affected user flow; passing a small unit test does not prove the whole product works.
The official permissions guidance distinguishes sandbox access from approval policy. A blocked command may need a narrower permission or environment fix, not a global removal of protections. Give the task only the files and services it needs, and keep production secrets out of sample prompts.
If comparing offers on FamilyPro, confirm whether the specific offer supports your intended Codex interface, whose account is used, and how usage and privacy are handled. A ChatGPT access offer alone is not proof of Codex eligibility or a private development workspace. Use an account and workspace you control for confidential repositories.
The current FamilyPro GPT product rules explicitly exclude Codex from shared accounts; Codex quota and secondary-verification issues are excluded from warranty and refunds.
Frequently asked questions
Is Codex free?
There is a limited Free-plan route in the current official pricing, subject to availability. Check your account instead of assuming all interfaces and models are free.
Do I need to copy a plan from ChatGPT into Codex?
Not necessarily. You can ask Codex to inspect the project, plan a bounded change, and implement it in the same workflow. Clear requirements matter more than moving text between products.
Will Codex deploy my application automatically?
That depends on the task, connected tools, and permissions. State deployment boundaries explicitly and review the change before authorizing publication. Generated code is not a production-readiness guarantee.
