How Codex's 5-hour and weekly limits work

OpenAI's help article "Understanding ChatGPT and Codex usage limits", read on 4 November 2026 and again on 5 November, announces the next change: "Starting in December, Plus, Pro, and Business plans' usage allowance will be shared across ChatGPT and Codex as we bring the agentic capabilities of Work into the main ChatGPT experience." Usage "will be measured in tokens", so how much a task uses depends on the model and the task, and "Instant chat will not draw from your usage." OpenAI expects "most users won't need to wait for their usage to reset more often than they do now", and says the combined remaining usage and reset times will show in Settings > Usage.

Part of that sharing already exists. OpenAI's article "Using Codex with your ChatGPT plan" says that on plans with shared allowances or credits, "Codex, ChatGPT Work, ChatGPT for Excel, and Workspace Agents use a shared allowance and credit pool". So from December, a long chat in ChatGPT can use up the same allowance as a Codex session, which makes a per-message table like the one below a weaker guide than it was.

OpenAI's Codex pricing page is the current statement. It gives a table of "local messages per five-hour period for Plus and Standard Business", by model, and the figures are ranges, because "The number of messages you can send depends on the model used, size and complexity of your tasks, and whether you run them locally or in the cloud." The page carries no date. Its table on 5 November 2026, unchanged since 3 November:

Under the table: "Local messages and cloud chats share your plan's usage allowance. Weekly limits may also apply." and "Cloud tasks may use more of your allowance than local messages." A local message is a turn in the Codex CLI or the IDE extension on your machine; a cloud task runs on OpenAI's side and counts for more.

Pro is the exception, in the page's words: "Pro plans currently have no five-hour limit." "Currently" is doing work there, as the next section explains. Enterprise and Edu are different again: "users with flexible pricing have no fixed rate limits. Usage scales with credits." For a Business workspace, OpenAI's enterprise usage-limits page says the workspace's usage limits and spend controls "apply to eligible activity under your workspace plan, which can include some Codex activity", but "don't cover all Codex usage or govern OpenAI API Platform billing".

How the window behaves is visible in the Codex CLI's own source rather than on a page. Each limit is a RateLimitWindow with used_percent, "Percentage (0-100) of the window that has been consumed", window_minutes, "Rolling window duration, in minutes", and resets_at, the time the window resets (openai/codex, protocol.rs). A snapshot holds a primary and a secondary window, which is how the CLI shows a 5-hour and a weekly limit side by side, or a weekly one alone.

The claim that OpenAI removed the 5-hour limit is true, dated, and only half the story. On 28 July 2026, OpenAI_Support replied on OpenAI's community forum to a user whose /status showed only a weekly limit: "As announced on July 12, 2026, the 5-hour Codex usage limit has been temporarily lifted for Plus, Pro, and Business users." That is the only statement from OpenAI itself that records the change.

On 24 August 2026, Tibo Sottiaux of OpenAI posted on X, as quoted by 9to5mac the same day: "Tomorrow we will bring back the 5h limit for Plus accounts across ChatGPT Work and Codex", and "We are for the upcoming months keeping the 5h limit not enabled for Pro $100 and Pro $200 subscriptions." That matches the pricing page today: Plus has the window, Pro does not. The post did not mention Business; the pricing page lists Standard Business beside Plus in the 5-hour table, so Business Standard seats have it again too.

Neither change appears in OpenAI's Codex changelog, which runs from May 2025 to October 2026 without an entry on limits. So a page or a forum post written between 12 July and 25 August that says "Codex has no 5-hour limit" was right when written and is wrong for Plus now. The word "currently" on the pricing page is the only hint that Pro's exemption is also a decision rather than a design, and "for the upcoming months" in August is the only duration anyone has given it.

OpenAI says "Weekly limits may also apply" and does not say how big they are, for any plan. The forum thread above shows what the CLI prints for a Pro account with no 5-hour window: a single line, "Weekly limit", with a percentage left and a reset time. On Plus, the two windows run at the same time, and the 5-hour one resetting does not give back any of the week.

Some Plus and Pro referral offers add a banked reset. In OpenAI's words, "Using a full banked reset refreshes your 5-hour and weekly Codex usage windows and changes your weekly reset date" (Using Codex with your ChatGPT plan). OpenAI Support does not reset limits on request.

The pricing page: "Credits are the unit used to pay for eligible usage on credit-based plans. After you reach your included limits, available credits let you continue working." The page prices models in credits per million tokens, with cached input far cheaper than fresh input, and does not give a dollar price per credit. So Codex's overage is a prepaid balance in credits that the limit falls through to, rather than a bill at the end of the month as Anthropic's usage credits are. On a Business workspace, the admin's usage limits and spend controls govern the eligible part of that, and "exhausted limits can pause access to eligible features". OpenAI's help centre adds two details: if you reach a limit in the middle of a turn, "Codex can continue working on that turn, subject to fair-use limits", and "Eligible Plus and Pro users can buy credits without changing plans" (Using Codex with your ChatGPT plan).

The pricing page answers this directly: "You can find your current limits in the usage dashboard. If you want to see your remaining limits during an active Codex CLI session, you can use /status." The dashboard is at chatgpt.com/codex/settings/usage, and the slash command reference describes /status as "Show the chat ID, context usage, and rate limits." The guide on how to check your Codex usage goes through both and the tools that keep a history.

Each seat has its own windows, and the Business admin pages set limits and spend controls for the workspace rather than show each person's window. For the team's question, who keeps running out and what it is costing, see what your team spends and who went past their plan; both cover Codex alongside Claude Code.

Codex writes each limits snapshot it receives to its session file on your machine. OpenReason Intel is a dashboard for Claude Code and Codex usage. Its reporter, or-otel, registers a Codex hook that runs after each turn, reads the newest snapshot, and sends only those fields: the 5-hour and weekly used percentage and reset times, credits, and the reason when a limit was hit. A seat with no 5-hour window is sent as weekly only. In the dashboard you can see your Codex windows over time, when each one filled, and the sessions and turns running when it did, beside the same view for Claude Code.

For one person, that adds a record across days and weeks and across every machine you use, where /status and the dashboard show only now; the turns that were running when a window filled, each with its estimated share; and your Claude Code windows beside your Codex ones. It does not warn you; it is a record you look at. What you can do instead is check /status during long sessions, and look afterwards at which turns filled the window. Sign-up takes a Google account or an email address, no card, and one command from the or-otel quickstart.