Limits & credits
Codex Task Limits 2026: Five-Hour Messages by Model, Cloud Tasks and Credits
OpenAI sets no Codex task count. It estimates 17–180 GPT-6.2 Sol and 6–50 GPT-6.2 Astra local messages per five hours on Plus and Business Standard, Pro removes the five-hour limit, and cloud tasks can use more.
Updated
In short
OpenAI publishes no Codex task limit for any plan. Each plan has a usage allowance, and OpenAI estimates how far it goes: on ChatGPT Plus, 17–180 local GPT-6.2 Sol messages and 6–50 GPT-6.2 Astra messages every five hours (alongside 35–285 on GPT-6.1 Terra, 15–160 on GPT-6.1 Sol and 5–45 on GPT-6 Astra), with cloud tasks able to use more and weekly limits that may also apply. Pro plans and Business Premium seats currently have no five-hour limit. Free and Go get Codex in the desktop app on GPT-6 Luna and GPT-5.6 Terra, subject to rollout, with no cloud tasks. Check remaining limits and reset times on the Codex usage dashboard or type /status in a CLI session.
On this page5
A task's size, not the number of tasks, decides how far the allowance goes
OpenAI says how many messages you can send depends on the model, the size and complexity of each task and whether it runs locally or in the cloud. Small scripts or routine functions may use only a fraction of the allowance, larger projects and sessions that hold more context use significantly more per message, and a long-running task can use substantially more than a short request. Cloud tasks draw on the same allowance as local work; standard Codex Cloud environments carry no separate machine charge at launch, so only the model's usage counts. Starting a new chat does not reset any limit, so the only task count worth planning with is your own: run a normal week and read the usage dashboard.
Which plans run local tasks, cloud tasks and GitHub reviews
Read across a row to see where a plan can run a Codex task and what stops it there. Among the ChatGPT plans, only Enterprise and Edu workspaces on flexible pricing have no fixed limit; every other plan stops at a five-hour window, the end of an included allowance or an empty credit balance.
Plan | Local tasks | Cloud tasks and GitHub reviews | Five-hour limit | Past the allowance |
|---|---|---|---|---|
Free and Go | Desktop app, on GPT-5.6 Terra and GPT-6 Luna (subject to rollout) | Not included | Not published | Credits for a limited group of users; otherwise wait or upgrade |
Plus | Desktop app, CLI, IDE extension, web and iOS (GPT-6.2 Astra & Sol included) | Included | OpenAI's per-model estimates | Credits |
Pro 100, 200 and 500 | As Plus ( | Included | None currently | Credits |
Business Standard seat | As Plus | Included | The same estimates; OpenAI's Business article gives | Workspace credits, if the owner bought them |
Business Premium seat | As Plus | Included | None | Workspace credits, if the owner bought them |
Legacy Business Codex seat | Codex only, without ChatGPT chat | As the workspace allows | No included allowance: every task is paid from workspace credits | Blocked when the workspace has no credits |
Enterprise and Edu | As the workspace allows (GPT-6.2 off by default until admin enables) | As the workspace allows; off in Enterprise until cloud access is enabled | With flexible pricing, no fixed rate limits; without it, the Plus limits for most features | The workspace's shared credit pool |
API key | CLI, SDK and IDE extension | Not available | No plan limit; billed at API rates | Pay as you go |
Pro 100 and Pro 500 include twice their October pre-increase Work and Codex allowance, Pro 200 sits above Pro 100 and adds Ultrafast for GPT-6.2 Sol, and Pro 500 adds the highest allowance plus Ultrafast for GPT-6.2 Astra (6x included usage; 3x credits through Feb 28, 2027, then 6x). OpenAI still cautions that a larger allowance does not guarantee twice as many completed tasks. A Premium seat gives five times the Standard Business allowance with no five-hour limit. The API key row is a separate bill rather than a plan; Codex ChatGPT login vs API key compares the two sign-ins.
GitHub reviews have their own counter; Work and, from December, chat share yours
A review that Codex runs through GitHub, when you tag @codex review on a pull request or turn on automatic reviews, counts as Code Review usage. A review you run locally with /review, or anywhere outside GitHub, counts toward the general limits like any other task, and so does building or editing a Site in Codex.
Codex shares that general allowance with ChatGPT Work, and from December so do chats on Plus, Pro and Business except Instant chat, with the combined usage measured in tokens. ChatGPT Work and Codex limits by plan covers everything else that draws on the pool.
When a limit hits in the middle of a task
Codex can finish the turn it is working on, subject to fair-use limits, and then stops. The limit notice or the usage page shows which allowance ran out, your credit balance and the reset time, and what your account can do next: add credits, upgrade or wait for the reset.
On Plus and Pro, credits carry the same task on once the included allowance is gone. A task that starts with a positive balance can end below zero if other usage drains it in the meantime, and a later purchase or automatic reload may first pay off that negative balance. In a Business workspace, a seat past its allowance continues only on credits the owner has bought; with none left, the feature is blocked until the owner adds more.
Fitting more tasks into five hours
OpenAI's advice is to shrink what each message carries: precise prompts without spare context, only the files the task needs, a defined output, a smaller AGENTS.md (nest the files per directory in a large repository), and fewer MCP servers, since every enabled server adds context to every message. Splitting a repository-wide change into smaller tasks keeps each message cheaper and lets you check /status between them. Routing routine edits to GPT-6.2 Sol (17–180 Plus messages) instead of GPT-6.2 Astra (6–50) gives roughly three times as many turns per five hours (GPT-6.2 Astra vs GPT-6.2 Sol), while GPT-6.1 Terra and GPT-6 Luna stretch the window further still. On GPT-6.2, Fast consumes the allowance at 2x the Standard rate (2.5x on older GPT-6/6.1 models) and Ultrafast at 6x (8x on 6 Astra/6.1 Sol), so leave speed tiers off for routine work.
Edited byRick Finch
Official sources
Editorial guidance grounded in official product sources.
- Codex official site
- Codex Pricing
- ChatGPT | ChatGPT Learn
- ChatGPT Pricing | OpenAI
- OpenAI introduces GPT-6 Sol and Luna
- OpenAI API pricing
- ChatGPT Work and Codex pricing
- About ChatGPT Pro tiers
Show all 19 sources
- Introducing GPT-6.1 Sol
- ChatGPT pricing
- Understanding ChatGPT and Codex usage limits (OpenAI Help Center)
- Using Codex with your ChatGPT plan (OpenAI Help Center)
- Using credits for flexible usage in ChatGPT, personal plans (OpenAI Help Center)
- ChatGPT Business models and limits (OpenAI Help Center)
- Flexible pricing for the Enterprise, Edu and Business plans (OpenAI Help Center)
- Managing legacy Codex seats in ChatGPT Business (OpenAI Help Center)
- Introducing GPT-6.2 Astra and Sol (OpenAI)
- GPT-6.2 release notes (ChatGPT Learn)
- Fast and Ultrafast modes (ChatGPT Learn)
FAQ
Common questions
Does free Codex access have task limits?
Yes. Free and Go include Codex in the ChatGPT desktop app on GPT-6 Luna, subject to rollout, without Codex Cloud, and OpenAI publishes no number for either plan. Credit purchases are open only to a limited group of Free and Go users, so most wait for the reset or upgrade when they reach the limit.
Do paid Codex plans remove task limits?
No. Plus and Standard Business seats have a five-hour window, and weekly limits may apply on any plan. Pro plans and Business Premium seats have no five-hour limit but still have an included allowance. Only Enterprise and Edu workspaces on flexible pricing have no fixed rate limits; their usage scales with credits.
Is there a fixed number of Codex tasks per month?
No. OpenAI sets a usage allowance, not a task count, and publishes only estimated local messages per five hours for each model. Model choice, task size, local or cloud execution and context all change how far the allowance goes, so measure a normal week of your own work on the usage dashboard.
Does an API key bypass ChatGPT Codex limits?
It moves the work to a different bill. Signed in with an API key, Codex runs in the CLI, SDK or IDE extension at standard API rates and outside your plan's limits, but without cloud tasks, GitHub code review or Slack. Any plan can run extra local chats this way after reaching its limit.
How should a Business workspace budget Codex use?
Start with Standard seats, which share the Plus estimates, and give Premium seats, which have no five-hour limit, to the heaviest users. Buy workspace credits only to keep people working past their seat allowance: without credits the feature is blocked at the limit. Legacy Codex seats bill every task to workspace credits and are open only to workspaces that had one before June 24, 2026 or a pending invitation that day.
Where can I see my remaining Codex limits?
On the Codex usage dashboard, which shows which allowance ran out, your credit balance and the reset time. In an active Codex CLI session, type `/status`. From December, Settings → Usage in ChatGPT shows the combined ChatGPT and Codex usage.
Next steps
Take the next buying step
- Tool profileCheck Codex pricingUse this next for the current Codex plan table, local-message ranges, credit rates, API boundaries, and workspace pricing details.
- GuideCheck free Codex accessUse this if the remaining question is whether Free or another included route is enough before you move into paid limits or credits.