Understand how Codex works across local and cloud workflows, how plan limits differ from API token billing, and how to control context, cost, and permissions.
Start with the essentials
OpenAI Codex is a coding agent that can investigate repositories, edit files, run commands, review changes, and carry work across local, IDE, cloud, and automated workflows. Codex usage can be included in an eligible ChatGPT plan or billed through APIs, depending on how you access it.
This guide was checked against official OpenAI documentation on October 21, 2026. Product models, allowances, and API prices change, so confirm the linked sources for your account before making a purchasing decision.
OpenAI's Codex documentation hub links current guides for the app, IDE, CLI, reviews, automation, and production workflows.
| Access method | How usage works | Where to check |
|---|---|---|
| Eligible ChatGPT plan | Included allowance or shared credits, varying by plan | Settings, usage dashboard, or /status |
| Codex with your API key | Model input, cached input, output, and tool charges | OpenAI Platform usage and billing |
| Agents API | Selected model rates plus applicable tools and sandbox compute | Platform usage and API response data |
| Enterprise token agreement | USD token billing under workspace controls | Workspace administrator and usage dashboard |
OpenAI notes that Codex task usage depends on the model, execution location, task complexity, context, reasoning level, speed, and tools. A long repository task can consume substantially more than a short request. Current plan limits are account-specific, so a static article should not claim an exact number of tasks. See Using Codex with your ChatGPT plan.
API pricing counts input tokens, cached input tokens, and output tokens separately. Built-in tool calls and hosted containers can add charges. OpenAI's current pricing page lists the active Codex models and rates; model-specific pages also show context windows and rate limits.
For a concrete calculation, if a Codex model costs $1.75 per million input tokens, $0.175 per million cached input tokens, and $14 per million output tokens, then a request with 100,000 uncached input tokens and 10,000 output tokens costs about $0.315: $0.175 input plus $0.14 output. Repeated agent turns, tool output, web searches, and container time can increase the full task cost. Check OpenAI API pricing for the current model table.
A model's context window is the maximum context it can process in one request. It is not the amount included in a subscription. Codex can compact previous work to preserve useful state while controlling context growth, but the most reliable saving is a focused task with explicit boundaries.
/status or the appropriate dashboard before starting another large task.Reproduce the reported pagination bug on the article listing. Inspect before editing. Preserve existing URLs and canonical behavior. Implement the smallest fix, then run the focused test, type checker, and production build. Report the root cause, changed files, command results, and anything that still requires browser verification.
This supplies a trigger, constraints, evidence, and a stopping condition. It gives the agent room to investigate while limiting unnecessary context and edits.
Agent-generated code can access the files, credentials, and network that its environment exposes. Use isolated environments for untrusted or multi-user workloads, restrict outbound hosts, and keep application credentials outside the agent environment where possible. OpenAI's sandbox security guide recommends isolation, restricted network access, separate environment credentials, and brokering third-party secrets.
ChatGPT is useful for explanation, brainstorming, and code snippets. Codex adds an agent harness around a model: it can gather repository context, use tools, edit files, run commands, follow permission policies, and keep working through failures. The underlying model matters, but the harness and environment determine what actions are possible.
Compare Claude Code for another terminal-centered agent, Cursor for an AI-first editor, GitHub Copilot for GitHub-native workflows, and Gemini Code Assist for Google Cloud teams. Our best AI for coding guide compares the workflows directly.
Availability and included usage depend on the current ChatGPT plan or workspace. API-key use is billed according to the selected model and tools. Check the live plan and pricing pages.
In supported Codex sessions, use /status. You can also inspect the ChatGPT usage view or OpenAI Platform usage dashboard, depending on how you authenticated.
There is no universal fixed cost per prompt. A task can involve multiple turns, model calls, tools, and growing context. Plan access uses allowances or credits; API access uses token and tool pricing.
Yes, when the selected Codex surface and permission policy provide a working environment. Review the commands and results rather than assuming a reported success is sufficient.
Prepared with AI assistance and reviewed against the sources cited in the article. Product details may change after publication.
Explore related topics
Software Engineer with 5+ years of experience specializing in scalable web systems, modern full-stack architecture, and technical research.
Continue reading
Artificial IntelligenceLearn Claude Code setup, model choices, context management, subscription versus API billing, token costs, and practical ways to reduce usage safely.
Artificial IntelligenceUnderstand Cursor plans, model context windows, token-based usage, Agent workflows, privacy controls, and how to estimate the cost of real coding tasks.
Artificial IntelligenceLearn how GitHub Copilot agents work, compare current plans and AI credits, understand token-based usage, and follow a practical setup and evaluation workflow.