Codex usage limit: 15 to 160 Sol messages per 5 hours on Plus, and what the wall message offers each plan

What is the Codex usage limit on each ChatGPT plan?

OpenAI gives Codex a rolling 5 hour window and a weekly cap on Plus and Business, and it prints a range, not a count. The ranges below are the "local messages per five-hour period" on OpenAI's Codex pricing page, read on 18 October 2026, and they're identical for Plus and a Standard Business seat. The spread is wide because OpenAI meters tokens. A one line fix and a refactor across 40 files both count as one message.

GPT-6 Luna sits off that scale at 350 to 3,000 messages per window, which makes it the model to switch to when the window is nearly gone. OpenAI's own caveat sits under the table: "These estimates are not fixed message limits." Two more lines on the same page matter: "Pro plans currently have no five-hour limit," and "Weekly limits may also apply." OpenAI publishes no weekly number for any plan, Pro included.

We compiled the table from OpenAI's Codex pricing page on 18 October 2026. The Pro multiples come from the press. Heise reported on 10 April 2026, citing OpenAI, that the $100 tier carries up to five times the Codex usage of Plus. OpenAI announced on 29 September that Pro $200 will fall from 20 times Plus to 10 times in Codex and Work from 30 October 2026, per TheNextWeb. Subscribers who hold it before the change keep 20 times through 29 October. They also get a one time grant of usage credits worth $2,500, which expires at the end of the year. The $500 tier is the only one with Ultrafast, which Engadget says carries OpenAI's highest allowance.

What does the Codex limit message tell you on each plan?

Every personal Codex limit message opens with "You've hit your usage limit," and what follows depends on your plan. Workspace caps open with "Your workspace is out of credits" or "You hit your spend cap" instead. We read the message builder in the open source openai/codex repository, file codex-rs/protocol/src/error.rs, on 18 October 2026. It branches on plan type, and the remedies differ more than OpenAI's pricing page suggests.

Each personal message then ends with a clock: "try again at" a time (with the date if it isn't today), or "try again later" when the server sent no reset time. Usage based Enterprise workspaces get the admin line, not the bare one. That time is the honest answer to when Codex comes back. On a Business seat you can't buy your way past the wall. Only an admin can. Enterprise and Edu seats without flexible pricing run on Plus limits, so a bare message there means that seat's window or weekly cap is spent.

One look-alike isn't a limit at all. "Selected model is at capacity" means OpenAI is short of servers for that model, and Codex selected model is at capacity covers that error. The desktop app and ChatGPT may word the limit differently. The CLI is simply the one place the logic is public.

How does the Codex 5 hour limit work with the weekly limit?

The 5 hour window and the weekly cap run at the same time, and whichever empties first stops you. Both count tokens. So you can hold weekly room and still be shut out until the 5 hour window refills. Or you can spread work across days and hit the weekly cap with the window full. Four things drain them faster than your message count suggests.

1. Cloud tasks. OpenAI says "Cloud tasks may use more of your allowance than local messages," and both draw on one pool.
2. Fast and Ultrafast. Fast mode spends included usage at 2.5 times the Standard rate, and Ultrafast on GPT-6 Astra or GPT-6.1 Sol at 8 times.
3. Images. Image generation uses included limits about 3 to 5 times faster on average, per the same page.
4. Context. Every file, AGENTS.md block and MCP server you load adds to the context Codex sends. OpenAI's own tips are to "Disable MCP servers when you don't need them" and to nest AGENTS.md files so less is injected.

If you run out partway through a turn, it isn't lost. OpenAI says the agent "will be able to continue working on that turn," subject to fair use, and the wall applies to the next one.

The window itself has moved twice this year. OpenAI's Codex lead, Thibault Sottiaux, lifted it for Plus, Business and Pro on 12 July 2026, leaving only the weekly cap. On 24 August he posted, "Tomorrow we will bring back the 5h limit for Plus accounts across ChatGPT Work and Codex," as 9to5Mac reported. It came back on 25 August for Plus, while Sottiaux kept it off for Pro $100 and Pro $200 "for the upcoming months," which is why Pro still has only a weekly cap.

Why did ChatGPT say you've reached your 5 hour Work usage limit?

ChatGPT Work spends the same allowance as Codex, so the 5 hour Work limit and the Codex 5 hour limit are one window seen from two apps. Same pool, two doors. OpenAI's pricing page says it plainly: "ChatGPT Work and Codex share usage." A long Work task in the afternoon leaves less for a Codex session that evening, and the reverse is true too.

So the fixes are the same ones. Wait for the window, switch the Work task to GPT-6 Luna, or buy credits on Plus. Three other things draw from that pool, per the same page: ChatGPT for Excel on Plus and Pro, image generation, and voice in the desktop app. OpenAI bills that voice at $0.05 a minute against your Codex budget. It doesn't list ordinary ChatGPT chat among them.

How do you check your remaining Codex usage?

Your remaining Codex usage is on the dashboard at chatgpt.com/codex/settings/usage, and in the CLI you type /status during a session. OpenAI suggests you "check the dashboard every week or two" to track your pace. In /status, each window shows as its own row, such as "5h limit", with a percent left.

The CLI also warns you before the wall. Its code, read on 18 October 2026, prints "Heads up, you have less than" a percent "of your" window "limit left" once you've used 50%, 75%, 90% and 95%. At 90% it offers to switch you to a smaller model "for lower credit usage". Take that offer when the job is mostly reading and editing files. Keep the bigger model for the one hard step.

What can you do when you hit the Codex usage limit?

You have five ways past the Codex usage limit, and the cheapest is usually the wait. Run them in this order.

1. Read the retry time. The message ends with "try again at" a time. That's when the window refills.
2. Check for a reset. Sottiaux has reset everyone's paid Codex usage often: codexresets.com, a tracker that logs his posts on X, counted 7 resets between 1 and 17 October 2026, one of them banked. The CLI also has a menu to apply a banked "Full reset" when your account holds one.
3. Switch models. GPT-6 Luna's range is 350 to 3,000 messages per window on Plus, against 15 to 160 for GPT-6.1 Sol.
4. Buy credits. On Plus and Pro, OpenAI says you can "purchase additional credits to continue working" without upgrading. It doesn't print one price, since "Credit purchase prices and applicable discounts depend on your plan or agreement." OpenAI's page puts a typical GPT-5.6 Sol task at 2 to 15 credits.
5. Use an API key. OpenAI says "All users may also run extra local chats using an API key," billed at standard API rates. That works in the CLI, SDK and IDE extension, with no cloud tasks or code review.

Upgrading from Plus to Pro $100 removes the 5 hour window today and adds up to five times the usage, though the weekly cap stays. If only one week a month hits the wall, price that week's credits in your usage settings first, since OpenAI prints no single credit price, and compare it with the $80 a month gap. ChatGPT Plus vs Pro prices that choice out. On a Business seat, the request goes to your admin, who can buy workspace credits. ChatGPT Business vs Plus covers what a Standard seat does and doesn't add.

Whizi is a different tool for the wait. It's a chat app, not a coding agent: it won't run in your terminal or edit your repo. What it does is keep you asking code questions on Claude and GPT while the window refills. Paste the file, get the fix. Whizi Pro is $29.99 a month for 2,500 credits, Claude Sonnet costs 10 credits a message, so that's 250 Sonnet answers, and Auto sends code questions to Claude Sonnet on its own (how Whizi credits work). If you need the agent itself, the fix is credits or Pro, not another chat app.