Codex Task Limits Explained: Free, Paid, Credits, and API Billing

Short answer: Codex task limits are not a fixed number of finished tickets. They depend on your ChatGPT plan or API route, the model, how much code and context the task needs, whether the work runs locally or through a workspace/cloud surface, and whether you keep going with credits after included usage runs out. Free access is useful for quick tasks, but paid access still has limits, and API-key usage is a separate billing path.

Use Codex Pricing for the current rate table and plan details. Use Is Codex Free? only if your main question is whether you can start without paying. This page is about the limit boundary: why one Codex task can be cheap, why another can burn through allowance, and who owns the usage when a team, CLI session, desktop app, or API key is involved.

Why task limits are not a fixed task count

OpenAI describes Codex limits as plan-based and task-dependent. A small script or routine function may use only part of an allowance, while a larger codebase, long-running implementation, or extended session needs more context and consumes more per message. That means a buyer should not compare Codex plans by asking only how many prompts are included.

For paid individual local-message usage, OpenAI's published Codex pricing table gives ranges rather than one exact number. GPT-5.5 local messages are listed at 15-80 per five-hour window on Plus, 80-400 on Pro 5x, and 300-1600 on Pro 20x. GPT-5.4 mini has higher listed ranges for routine local messages, while API-key usage is shown as usage-based rather than a five-hour local-message allotment.

Treat those ranges as planning signals, not a guarantee that every task has the same weight. Model choice, repository size, prompt length, retained context, MCP servers, retries, generated output, and execution surface all change how much allowance a task consumes. The practical question is how often Codex will do real repo work for you, not whether one example prompt fits inside a headline range.

How credits change the ceiling

Credits are the bridge after included usage. OpenAI says Plus and Pro users who reach Codex limits can add credits, while Free and Go users are prompted to upgrade to Plus instead of adding Codex credits. On Business, Edu, and Enterprise plans with flexible pricing, workspace credits can extend usage for the organization.

The credit model is token-based. OpenAI's Codex rate card maps usage to credits per million input tokens, cached input tokens, and output tokens, and it notes that a typical GPT-5.5 Codex task may consume 5-45 credits. That range explains why a lightweight local fix and a broad migration do not feel like the same billing event even if both start as one user prompt.

Some settings and features spend allowance faster. Fast mode increases speed by consuming more credits for supported models. Image generation counts toward general Codex usage limits and can draw down included limits faster than similar turns without image generation. If you are trying to stretch limits, smaller models, tighter prompts, and less unnecessary context matter.

CLI and desktop expectations

The Codex app, CLI, IDE extension, and web/cloud surfaces are access surfaces, not separate universal pricing plans. The desktop app is available on macOS and Windows, the CLI runs from a terminal, and both can be used for local project work after authentication. What you sign in with determines the billing and governance boundary.

With ChatGPT sign-in, Codex uses your ChatGPT account or workspace entitlement. The CLI and IDE extension also support API-key authentication, but Codex cloud requires ChatGPT sign-in. The practical distinction is simple: a local CLI session can still be using ChatGPT plan limits, while an API-key session is billed through the OpenAI Platform account.

For active use, OpenAI points users to the Codex usage dashboard and says the CLI /status command can show remaining limits during a session. That is the right operational check because limits are dynamic: the same account may have plan allowance, purchased credits, workspace credits, or API usage depending on how the current session is authenticated.

Business and workspace ownership

Business buyers should separate individual plan limits from workspace ownership. ChatGPT Business has standard ChatGPT seats that include ChatGPT and Codex, and Codex seats that provide Codex-only access on a usage-based model. A Codex seat is not just a cheaper personal subscription; it changes who owns access, billing, and admin controls.

Workspace credits are also an admin problem, not only a developer problem. OpenAI describes automatic reload, monthly recharge limits, credit maximums by seat type, and per-user overrides for Business workspaces. That lets a team give different limits to Codex-heavy developers without treating every user as the same kind of ChatGPT seat.

For larger organizations, workspace analytics and audit surfaces matter. OpenAI's Codex governance docs describe usage tracking across surfaces such as CLI, IDE extension, cloud, desktop, and Code Review, with credit and token usage by product surface or model. Those controls are often the reason to buy through a workspace instead of letting each developer use a personal plan.

When API billing becomes separate

API billing becomes separate when Codex is authenticated with an OpenAI API key or when your own software calls OpenAI models directly. OpenAI's authentication docs say API-key Codex usage follows standard API pricing instead of included ChatGPT plan credits, and the API pricing page says API access is billed separately from ChatGPT Plus, Business, Enterprise, and Edu.

That API boundary is especially important for CI, scripts, SDK workflows, internal tools, or product features. API-key Codex can be useful for local automation in shared environments, but it does not include every cloud or workspace feature tied to ChatGPT-based Codex access. Budget API usage as a developer-platform meter, not as a hidden part of a ChatGPT subscription.

The clean buying rule is to name the owner before naming the limit. If a person is using Codex in the app, CLI, IDE, or web through ChatGPT, start with plan limits and credits. If a company owns access for developers, start with workspace seats, credits, policies, and spend controls. If software owns the work through an API key, use API token pricing and separate API budgets.