In this guide:
- What OpenAI actually confirmed about the Codex limits bug
- How Codex usage limits actually work
- How to check your real Codex usage (60 seconds)
- False limit or real limit? The 6 signs
- When does the Codex usage limit reset?
- What to do while you wait for quota
- Lookalike errors that are NOT your usage limit
- Frequently asked questions
Your Codex usage limit just ran out — again — and you have barely done any real work. Before you blame your plan, read this: on October 20, 2026, OpenAI publicly confirmed that a bug applied the wrong usage limits to approximately 40,000 Codex accounts. If your meter has been draining at a speed that makes no sense this week, you may not have hit a real cap at all.
This guide separates the two situations. You will learn exactly what OpenAI admitted, how Codex limits actually work on each plan, how to check your real usage in under a minute, the six signs that your limit error is the bug rather than your own usage, when your limit resets, and what to do while you wait for quota to come back.
Update: October 28, 2026. Re-checked against OpenAI's Help Center on October 28. The estimates table, the shared Work and Codex allowance, the five-hour and weekly windows, and the rule that Pro 100, Pro 200 and Pro 500 currently have no five-hour limit still match the Help Center article. The Help Center also confirms that Fast mode uses included allowance at 2.5 times the Standard rate and Ultrafast at 8 times, that Support cannot reset limits, and that eligible Pro 200 subscribers keep the previous allowance through October 29, 2026 and then move to the lower allowance. The October 20 statement about approximately 40,000 accounts and any credits update could not be verified from an OpenAI page in this check, and no follow-up statement appeared in the Help Center pages checked.
What OpenAI actually confirmed about the Codex limits bug
Reports started over the weekend of October 17–18, 2026: subscribers said their Codex quota was draining far faster than the same work had cost a week earlier. On October 20, Tibo Sottiaux — the OpenAI lead for Codex and ChatGPT — answered publicly on X:
On October 20, Tibo Sottiaux, the OpenAI lead for Codex and ChatGPT, said in a public post that reports of usage draining faster than expected over the weekend had led OpenAI to identify approximately 40,000 accounts where the usage limits applied did not match the plan the user had selected, that OpenAI had corrected the issue, was reaching out to affected users individually, and would add credits to affected accounts. The exact wording of that post was not re-verified for this update.
Three things in that statement matter for you:
- The bug is fixed on OpenAI's side. The wrong limits are no longer being applied.
- OpenAI is contacting affected users individually — you do not need to file a claim to be on the list.
- Affected accounts get credits added. OpenAI has not said how many credits, or exactly when they land.
What OpenAI has not said is just as important. It has not named the plans involved, has not published a way to check whether your account is one of the 40,000, and has not linked this bug to every fast-drain report — including the separate complaints about GPT-6.1 Sol usage. So the honest position is: some limit errors this week are the confirmed bug, some are real caps, and the checks below are how you tell them apart.
How Codex usage limits actually work
Codex does not have a fixed "messages per day" number. OpenAI's own Help Center article is explicit: the ranges "are estimates, not guaranteed message counts," because what a message costs depends on the model, the size and complexity of the task, how much context the agent holds, reasoning effort, tool use, and whether the task runs locally or in the cloud. Two more facts shape everything:
- Codex and ChatGPT Work share one allowance. Heavy Work sessions drain the same pool your coding uses.
- Local messages and cloud chats share the pool too — and cloud tasks usually cost more per task than local messages.
The two meters: a 5-hour window and a weekly cap
On Plus and Standard Business, usage is metered in a rolling five-hour window, and a weekly limit may also apply on top. OpenAI's published estimates for local messages per five-hour window on Plus are:
| Model | Estimated local messages per 5 hours (Plus) |
|---|---|
| GPT-6 Astra | 5–45 |
| GPT-6.1 Sol | 15–160 |
| GPT-6 Sol | 15–150 |
| GPT-6 Luna | 350–3,000 |
On Pro plans there is currently no five-hour limit — only the weekly allowance gates you. That is why a Pro user who suddenly hits a wall mid-afternoon should look at the weekly meter first. (The five-hour window itself is not new trouble: OpenAI restored it for Plus back on August 25, 2026, after a temporary removal, and said so publicly at the time.)
What burns your allowance faster than you expect
- Speed modes. Fast mode uses your included usage at 2.5× the Standard rate; Ultrafast on GPT-6 Astra or GPT-6.1 Sol uses it at 8×.
- Image generation uses included limits roughly 3–5× faster than a similar text-only turn.
- Long sessions with big context. The more of your codebase the agent holds, the more each message costs.
- Extra baggage on every message. A large AGENTS.md file and every enabled MCP server add context — and cost — to each request.
How to check your real Codex usage (60 seconds)
Do not guess from memory. OpenAI gives you two official places to look, and they should agree with each other:
Option 1: The usage dashboard (works everywhere)
- Go to chatgpt.com/codex/settings/usage — or open ChatGPT Settings → Usage.
- Read the meters for the current window and the week. Note the percentage used and the reset time shown next to each limit.
- If you use the Codex app, the same information lives under Settings → Usage & billing, including a usage analytics breakdown by model and by chat.
Option 2: /status inside the Codex CLI
- Open a Codex CLI session (or use the CLI built into the desktop app).
- Type
/statusand press Enter. - Codex shows your remaining limits for the active session — run it before a long task, not after the task stalls.
OpenAI's own advice is to check the dashboard every week or two so you learn your normal pace. That baseline is what makes an abnormal drain obvious when it happens.
False limit or real limit? The 6 signs
Work through this checklist against what your dashboard shows. The more "bug" signs match, the less likely your own usage explains the wall.
Signs it is probably the bug (or a counting error)
- The percentage jumped in a huge leap. In one public GitHub report from October 16, a Pro account's weekly meter went from 19% to 75% in about 15 minutes of ordinary work. Real usage climbs; it does not teleport.
- Usage is attributed to a model you never selected. One Pro user found roughly 80% of a week's usage charged to GPT-6 Astra in the analytics view while every individual chat showed GPT-6.1 Sol. If the by-model chart and the per-chat breakdown disagree, the counting is wrong somewhere.
- Two official meters contradict each other. Users have been refused Codex code reviews for "usage limits" while the usage page showed half the weekly allowance still remaining. Contradictory meters are an OpenAI-side problem, not yours.
- The drain started over the October 17–18 weekend with no change in how you work — that is the exact window OpenAI's statement covers.
Signs it is probably a real limit
- Your settings explain it. You have been running Fast or Ultrafast mode, maximum reasoning, image generation, or very long high-context sessions. At 2.5×–8× burn rates, a real cap arrives fast.
- You are a grandfathered Pro $200 subscriber approaching October 30. OpenAI's Help Center says grandfathering applies if you had an active Pro 200 subscription at any point from September 22, 2026 through 10 a.m. Pacific Time on September 29, 2026, and eligible customers can use the previous included allowance through October 29, 2026 while their Pro 200 subscription is active, then move to the lower included usage allowance. A smaller allowance after that date is a published plan change — expected, not a bug.
If signs 1–4 fit you, the practical move is patience plus evidence: screenshot your dashboard meters (percentages and reset times), note the dates, and watch for OpenAI's individual outreach and the promised credits. If you contact Support, those screenshots and timestamps are what a counting investigation needs.
When does the Codex usage limit reset?
- The 5-hour window (Plus and Business) is rolling: usage frees up five hours after it was spent, not at midnight. A heavy 9 a.m. stretch has cleared by 2 p.m.
- The weekly limit resets on an account-specific date and time — yours is shown in the usage dashboard next to the weekly meter. There is no single global reset moment for everyone.
- Banked and purchased resets can move your reset earlier; if you have a saved reset available, it refreshes the limits it covers when used.
- OpenAI-wide resets happen occasionally — whether an all-subscriber reset also landed on the morning of October 20 alongside the bug statement could not be confirmed from the OpenAI Help Center pages checked for this update. They are goodwill gestures, not a schedule you can count on.
- Support cannot reset your limit for you. OpenAI's help guidance is blunt on this point — asking support for a manual reset will not work. Your real options are below.
What to do while you wait for quota
- Let the current turn finish. If you hit the limit mid-task, OpenAI lets the agent complete the turn already in progress, subject to fair use. Do not kill a nearly-done task.
- Switch to a cheaper model. GPT-6 Luna's allowance stretches dramatically further than Astra or Sol on the same plan. Use the big models for the work that needs them and Luna for routine edits.
- Drop back to Standard speed and sane reasoning. Fast at 2.5× and Ultrafast at 8× are how allowances evaporate. Save them for deadlines.
- Trim what rides along. Shrink oversized AGENTS.md files, disable MCP servers you are not using, and start fresh chats instead of dragging a giant context through every message.
- Buy credits if the work cannot wait. Plus and Pro users who reach their limit can purchase additional ChatGPT credits and keep working on the same plan — no upgrade needed.
- Run local chats on an API key. Any user can run extra local Codex chats authenticated with an OpenAI API key, billed at standard API rates, completely separate from the subscription allowance. You give up cloud features (GitHub code review, Slack) on those sessions.
Lookalike errors that are NOT your usage limit
- "Selected model is at capacity. Please try a different model." That is OpenAI-side serving capacity for that model in your region — not your quota. Your remaining usage does not affect it. Wait a few minutes or switch models with
/model. - Codex stops you in regular Chat mode instead. Limit messages inside Codex and Work come from the shared agentic allowance; a plain Chat conversation does not draw on it. If Chat itself is erroring, that is a different problem — see our guide to the ChatGPT "Model Not Found" error and its 6 fixes.
- Privacy or account confusion while you troubleshoot. While you are in ChatGPT settings checking usage, it is also worth knowing about the newest account feature — our walkthrough of ChatGPT Locked Chats and the PIN reset warning covers what locking does and does not protect.
Frequently asked questions
What is the Codex usage limit bug OpenAI confirmed?
On October 20, 2026, OpenAI's Codex lead Tibo Sottiaux said approximately 40,000 accounts had an inconsistency where the usage limits applied did not match the plan the user had actually selected. OpenAI says it has corrected the issue, is reaching out to affected users individually, and will add credits to affected accounts.
How do I check my Codex usage limit?
Open the usage dashboard at chatgpt.com/codex/settings/usage (or ChatGPT Settings → Usage) to see your current limits and reset times. Inside an active Codex CLI session, type /status to see your remaining limits without leaving the terminal.
When does my Codex usage limit reset?
On Plus and Business, the five-hour window is rolling — usage frees up five hours after it was spent — and the weekly cap resets on the account-specific date shown in your usage dashboard. Pro plans currently have no five-hour limit, so the weekly reset is the one that matters.
Can OpenAI Support reset my Codex limit?
No. OpenAI's help guidance states that Support cannot reset ChatGPT or Codex limits. Your options are to wait for the reset shown in your dashboard, use a banked reset if you have one, buy additional credits on Plus or Pro, or run local chats with an API key.
Why is my Codex usage draining so fast if I am not one of the 40,000 accounts?
Because the meter counts work, not messages. Fast mode burns included usage at 2.5× and Ultrafast at 8× the Standard rate, image generation runs about 3–5× faster, cloud tasks cost more than local ones, and long sessions carrying a large context — plus big AGENTS.md files and extra MCP servers — raise the cost of every single message.
What should I do if I think the bug drained my quota?
Screenshot your usage dashboard showing the percentages, the by-model breakdown, and the reset times, and note when the abnormal drain happened. OpenAI says it is contacting affected users individually and adding credits, so keep that evidence in case you need to follow up with Support about a counting error.
Bottom line: a Codex limit that arrives after a weekend of normal work, in one giant percentage leap, is worth suspecting — OpenAI has admitted the counting was wrong for roughly 40,000 accounts. Check the dashboard, run the six-sign test, and stretch what you have left with cheaper models and Standard speed until your reset lands.
Taking a break from coding? See the Xbox Game Pass new games October 26–30 — 50 new titles arrive, and only 3 are on Game Pass.