Disclosure: we build Codex Reset. This post explains how its reset forecast works, including where it falls short. The details come from the site's public method page and its live forecast API.

The Codex Reset homepage answers one question first: will Codex reset today? It shows two numbers, the chance of a global reset in the next 24 hours and in the next 48 hours. People read those numbers in different ways. Some treat them as a countdown. Some assume the chance must be climbing because the last reset was a while ago. Neither is how they work.

The last post covered the 5-hour and weekly windows and the kinds of reset OpenAI describes. This one is about the global resets that arrive without a schedule, and how we put a number on them.

Two inputs, kept apart

The forecast has two sources, and it doesn't blend them.

The first is what Tibo (Thibault Sottiaux, @thsottiaux, who leads Codex) posts on X. Almost every global reset so far has been announced there. When he explicitly promises a future global reset, the site opens a ⚠️ Watch. It scores 93% when his words give a date, a range or a clock time, and 83% when they don't. Those are the only two scores. Hedged or suggestive wording can stay visible on the site as context, but it never sends a notification. Posts by other people can't create a Watch, and neither can a follow-up that only adds a time.

The second input is history. When no scored post is live, the site falls back to a cadence model built from the verified record of past resets. That record starts in September 2025 and held 60 global resets when this was written.

The cadence model

The current version is called rate-v4. It takes the last eight completed intervals between resets, plus the time since the most recent one, and estimates how often resets are happening right now. The 24-hour and 48-hour numbers come from that rate.

The part that surprises people is how it treats the current wait. In the method page's words: "The ongoing wait adds exposure without adding a reset, so the baseline decreases as waiting continues." Every quiet day is a day of watching with no reset. The model counts it as evidence that resets are rarer at the moment, not as a sign that one is overdue.

We tested the intuitive version too. The homepage states the result: "models that treated a reset as becoming 'due' scored worse in the backtest." So the number doesn't go up while you wait.

Here is what that looked like on November 5, 2026 at 15:06 UTC. The last verified reset was on October 28, 8.4 days earlier. All of the last eight completed intervals were shorter than that: the typical wait was 2 days, the average 2.1 and the longest 3.5. If waiting made a reset "due", this would be the moment for a high number. The forecast said 32% for the next 24 hours and 54% for the next 48. We read it again at 23:40 UTC the same day, after 8.7 days of waiting: the 24-hour number was still 32%, and the 48-hour number had slipped to 53%.

The forecast card on codex-reset.com (screenshot, November 5, 2026). The historical peak is shown in the visitor's local time; this browser was on GMT+8, so 12 PM–3 PM is 4–7 AM UTC.

Two more rules shape what you see. With no signal live, the page shows at most 46% for 24 hours and 71% for 48 hours; the model's own figures stay in the live JSON. And each percentage is rounded to a whole point and comes with a range: 17–52% for that 32%. The range shows how thin the evidence is. The method page is careful to add that it is "not a promise that the truth lands inside it."

"The cadence model, honestly" on codex-reset.com/forecast-method (screenshot, November 5, 2026). The boxed backtest note at the bottom is the page's August snapshot; the current scores are in the next sections.

What we tried and left out

When Tibo hints at a reset without promising one, it is tempting to nudge the number up. We tested that. The site still shows the state of his reset-related posts over the last seven days, with the post that decided it, but that state isn't counted in the percentage. A test on September 9, 2026 found that weighting the percentage by it made the forecast worse. It stays out until the record shows it predicts something.

At the snapshot above, the state was a tease from an October 31 post. The API marked it "effect": "informational_only" with a multiplier of 1.

How good is it?

The forecast is scored with a walk-forward backtest: each past prediction uses only the data that existed at that moment, and is then compared with what happened. The score is the Brier score, the average squared gap between the predicted probability and the outcome (1 if a reset came, 0 if it didn't). Lower is better.

At the snapshot, over 369 predictions:

• the live model: 0.130
• a naive baseline: 0.133
• the previous model, rate-v2: 0.129

So the live model beats the naive baseline and is slightly behind rate-v2. The differences are a few thousandths. The site says this openly: the API marks the forecast "confidence": "low" with the note "Experimental: walk-forward backtest has not beaten both baselines yet.", and the homepage labels it "Experimental · not above baseline".

The ⚠️ Watch scores have a separate record. The method page keeps a table from before the current alert rules, dated August 27, 2026, exactly as observed: dated commitments had resolved twice, with one hit, and the other bands are marked "insufficient history". Two cases are not a track record yet.

The time of day

Verified resets cluster in some hours. In the record, the most common window is 4 AM to 7 AM UTC. The method page calls this "a rhythm, not a schedule: historical clustering cannot name the next arrival." It tells you when to look, not whether a reset comes today.

What it can't tell you

• Your own limits. Codex Reset never reads anyone's OpenAI account, quota or usage. Your 5-hour and weekly windows are in Usage settings or /status.
• What OpenAI plans. A forecast "is not an OpenAI schedule, leak, or commitment, and it does not predict outages."
• Exactly what Tibo meant. His posts are read by an automated language model, which can get it wrong when he is vague.
• Whether a reset reached you. OpenAI says "'Global' does not mean every account or every usage limit."
• Hard bounds. The range beside a percentage "is not a calibrated coverage guarantee".

Reading the numbers

• With no ⚠️ Watch live, the 24-hour number is a base rate. A 32% day is more likely to end without a reset than with one, so if your weekly usage is gone, plan around the weekly rollover rather than a rescue.
• A ⚠️ Watch at 93% means Tibo gave a time. It is still a forecast.
• ✅ is different. It comes when the Pro and Plus accounts we monitor actually show the reset; Tibo's words alone never send one. If the monitors can't take a reading, a ✅ that Tibo has confirmed says "Our monitor hasn't measured it yet". The channels that deliver these are in the post on alerts.

If you want the inputs behind today's number, the full forecast is public JSON at /api/forecast, no key needed, with the method explained on the forecast method page.

Codex Reset is an independent community tracker and is not affiliated with OpenAI. "Codex" is a trademark of OpenAI.

Disclosure: we build Codex Reset. This post explains how Codex limits and resets work, using OpenAI's own pricing page and Help Center as the source, and points to the pages on our site that track them.

Most of the confusion around Codex limits comes from one word doing too many jobs. "Reset" can mean your 5-hour window rolling over, your weekly window rolling over, a reset OpenAI applies to many accounts at once, a reset saved in your account, or one you buy. They behave differently, and some of them move your next reset date.

The last post covered where to get reset alerts. This one is the background: what the windows are, what OpenAI does and doesn't publish about them, and what each kind of reset actually changes. Everything below comes from OpenAI's pages as of November 5, 2026. OpenAI updates them often, so treat the links as the authority and our summary as a map.

Two windows, one pool

On most paid plans Codex runs two limits at once: a 5-hour window and a weekly window. The same usage counts against both, but the two periods can end at different times. OpenAI's own example: your weekly period could end at noon while your current 5-hour period runs until 3 p.m.

That's why clearing one doesn't fix the other. If your weekly window is exhausted, a fresh 5-hour window won't unblock you. You need the weekly rollover, a reset, or credits.

The pool is also bigger than Codex. Where your plan has them, Codex, ChatGPT Work, ChatGPT for Excel and Workspace Agents share one allowance and credit pool, so a long ChatGPT Work session spends the budget your coding session needs. Local messages from the CLI, IDE extension or desktop app share it with cloud chats, and cloud tasks may use more of it than local messages.

Plans differ in which windows they have:

• Plus and Business Standard: a 5-hour allowance and a weekly allowance.
• Pro (100, 200 and 500): the pricing page says "Pro plans currently have no five-hour limit", so Pro runs on the weekly allowance.
• Go: a monthly allowance.
• Enterprise and Edu with flexible pricing: no fixed rate limits; usage scales with credits.

What the 5-hour numbers mean

The pricing page lists local messages per five-hour period for Plus and Standard Business. The ranges are wide: GPT-6.1 Sol 15–160, GPT-6.1 Terra 35–285, GPT-6 Astra 5–45, GPT-6 Luna 350–3,000.

OpenAI's caveat is the important part: "These estimates are not fixed message limits." What a message costs depends on the model, where the task runs, how complex it is, the context, reasoning, speed and tools. A long-running agent task can use far more than a short request. Read the ranges as a sense of scale, not as a counter.

The weekly window has no published size

For the weekly window OpenAI publishes no number for any plan. The pricing page says only "Weekly limits may also apply." Anyone quoting an exact weekly cap is estimating.

People do try to measure it. Our /codex-usage page collects those measurements, converts one full week into API dollars per plan and cites every figure. OpenAI warns against that kind of conversion: "API token prices are separate from subscription usage; don't use them to estimate included tasks." So we show the table as a rough scale, with each row's source next to it.

"Codex limits by plan, with sources" on codex-reset.com/codex-usage (screenshot, November 5, 2026). The dollar figures are third-party measurements, not OpenAI numbers.

Where to check your own numbers

Your own balance is the only number that counts. OpenAI points to these places:

• In an active Codex CLI session, run /status.
• On the web, ChatGPT Settings → Usage (chatgpt.com/settings/usage).
• In the desktop app, Settings → Usage & billing.

When you hit a limit, the notice shows the options your account has: add credits, use an available reset, upgrade, or wait until the displayed reset time.

Three kinds of reset

OpenAI's article How resets work in Codex separates three one-time resets, all distinct from the regular scheduled refresh:

• Banked reset. Free, one-time, saved to your account for later. Receiving one does not refresh your usage. You choose when to use it, before it expires.
• Automatic reset, "sometimes called a global reset". Applied directly to the accounts and limits an offer selects. You don't redeem it and it isn't saved. In OpenAI's words: "'Global' does not mean every account or every usage limit."
• Purchased reset. Bought for immediate use and offered to selected Plus and Pro personal accounts.

None of them permanently raises your limits. Free banked and automatic resets have, as OpenAI puts it, "no guaranteed availability or schedule". That uncertainty is what our site tracks: it reads the public announcements, shows a forecast, and confirms when a reset has actually reached the accounts we monitor.

Grant, redemption and automatic reset compared on codex-reset.com/banked-reset (screenshot, November 5, 2026).

Using a banked reset

You can redeem a banked reset in the desktop app (Settings → Usage & billing), on the web (Usage settings), or in Codex CLI by entering /usage and choosing Redeem reset. The mobile app doesn't offer it, and older CLI versions may not show the option.

A few rules catch people out:

• From Usage settings, at least one covered allowance must have less than 80% remaining. Otherwise you see "Your usage does not need a reset right now" and the reset stays in your account.
• A full banked reset brings each covered allowance that is below 100% back to 100%. An allowance already at 100% isn't restarted. On Plus that covers the 5-hour and weekly allowances; on Pro, the weekly allowance.
• It only counts as used if it restores at least one allowance.
• Separate allowances such as image generation, file upload and regular Chat Voice stay as they were.
• A desktop option, "Allow your assistant to use resets", lets Codex redeem one itself once your main 5-hour or weekly allowance is at 10% or less.

Use it before the expiry shown in your account or offer. Our banked reset page has an expiry tracker and a checklist for when a reset looks missing.

Why your weekly date moved

This is the most common surprise. Banked, global and paid resets can all change your next refresh time, and "the new schedule replaces the old one." OpenAI's example: a weekly period that was due to end on Friday restarts on Tuesday, so it now refreshes the following Tuesday, and the Friday refresh no longer applies.

So OpenAI's advice is to read the next refresh time in Usage instead of working it out from when you redeemed, bought, or saw an announcement. Before a new period starts, the displayed time can itself be an estimate.

Two related points from the same article:

• A global reset after a banked one brings the allowance back to 100%, not 200%. The banked reset you used isn't returned, and any unused ones keep their original expiry.
• The desktop History view shows banked-reset activity for the past 30 days. Paid and global resets don't appear there, so a missing entry doesn't mean you missed a global reset.

Paid resets

Eligible Plus and Pro personal accounts can buy an instant reset from desktop Usage settings, or from an offer that appears after reaching the weekly limit. It isn't available on Free, Go, Business, Enterprise or Edu, and availability varies by account even on the same plan.

A purchase restores both 5-hour and weekly usage at once. It pulls your normal weekly allowance forward rather than adding usage, can't be saved for later, and is generally not refunded. Your new weekly period starts with your first Work or Codex request after the reset, and the next automatic weekly reset is seven days after that request. It applies even if you still have usage left, and it changes your weekly reset schedule.

Where Codex Reset fits

Our site never reads your OpenAI account. The /codex-usage tracker lets you paste /status or type your weekly percentage and reset time, and keeps that in your browser. Its main job is the part OpenAI doesn't schedule: watching for automatic and banked resets and confirming when one has reached the Pro and Plus accounts we monitor. The alert channels are covered in the previous post.

What it can't do: it can't see your personal limits, it doesn't cover Team or Business workspaces, and it can't tell you whether a given offer included your account. For that, your own Usage page is the authority.

Quick answers

• My weekly allowance is empty. Does a fresh 5-hour window help? No.
• How big is the weekly limit? OpenAI doesn't publish it.
• Does receiving a banked reset refresh my usage? No. Only redeeming it does.
• Why did my weekly reset date change? A reset restarted the period. Check Usage for the new time.
• Does a global reset reach everyone? Not necessarily. OpenAI says "global" doesn't mean every account.

Codex Reset is an independent community tracker and is not affiliated with OpenAI. "Codex" is a trademark of OpenAI.

Disclosure: we build Codex Reset. This is a guide to its alert channels from the people who run it, including what they can't do.

If you use Codex heavily, you have probably watched your weekly limit drop to a few percent halfway through a task. OpenAI hands out global resets from time to time, and almost all of them are announced on X by Thibault Sottiaux (@thsottiaux, usually called Tibo), who leads Codex. So the default move is to keep refreshing his profile.

Codex Reset reads those public posts for you, checks Codex Pro and Plus accounts that we own to see when a refill has actually landed, and sends the result to the channel you pick. It never touches your OpenAI account, and you don't need an account with us. Earlier posts covered how we read Tibo's posts and how our own accounts confirm that a reset has landed.

This one is practical: where you can get the alerts, how quickly each channel tends to arrive, and the small catches.

What counts as an alert

Before picking a channel, it helps to know what will reach you. The alerts menu on the site sums it up in one line: "⚠️ strong Tibo signal · ✅ verified reset · 🎟️ banked reset — resets only, nothing else."

The alerts menu on codex-reset.com, open over the homepage (screenshot, November 4, 2026).

In practice that means a handful of message types:

• ⚠️ Watch: Tibo has made a clear promise. It comes with a probability: 93% when his words give a time window, 83% when they don't.
• ⌛️: Tibo says the reset is done, and our accounts haven't confirmed it yet.
• ✅: the refill reached the Pro and Plus accounts we monitor. There is one exception, covered under limits below, and the message says when it applies.
• 🔬: our accounts measured a reset that nobody announced.
• 🎟️: a banked reset update.

A real ✅ from the public Telegram channel is headed "✅ Codex reset confirmed" and reads "Our checks show that Codex usage limits reset around 06:35 UTC on Oct 28." Times are given in UTC.

Tibo's hints and teases (👀) are not alerts. They show on the site, and our X account posts them, but Telegram, Discord, email and push stay quiet. If you only want to hear about resets, any of the main channels will do that.

Telegram

t.me/codexresetalerts is the alerts channel. Open the link and join. It's a broadcast channel, so you only receive messages.

Telegram is one of the instant channels. It carries regular resets, strong Watches and qualifying banked updates. One detail makes it calmer than it sounds: when Tibo replies to his own post with a landing time, we edit the existing ⚠️ message with the new deadline instead of sending another one. You don't get pinged twice for the same reset.

Pick Telegram if it's already open on your phone or desktop all day.

Discord

The Discord invite takes you to the Codex Reset server and drops you in #alerts. That channel carries the same alerts as Telegram, and the same in-place edit applies when a landing time arrives.

Discord fits if your team or coding community already lives there. Discord lets you turn on notifications for #alerts and mute the rest of the server, which keeps it to resets only.

Email

Email starts at notify.codex-reset.com. Enter your address and we send a confirmation link. Opening the link doesn't subscribe you on its own. You have to press the confirm button on the page it opens.

The email subscription page at notify.codex-reset.com (screenshot, November 4, 2026).

The confirmation comes from alerts@notify.codex-reset.com. The sending domain is new, so check Spam if it doesn't show up. If it's there, marking it "not spam" helps later alerts reach your inbox.

After that you get one email for each confirmed reset, high-confidence forecast or qualifying banked-reset update. Two caveats are worth knowing:

• Email is the slowest lane. Messages go out in batches, and inbox providers add their own delay. An email can arrive minutes after the same alert is on Telegram, Discord or push.
• Delivery is at-most-once. In a rare, ambiguous failure at the email provider, you might miss one message. The system won't send a duplicate to make up for it.

The emails are transactional only: no sponsored content, affiliate placements, personalization or tracking. Each one has an unsubscribe link, and mail apps that support one-click unsubscribe will show that button too.

Email suits you if you'd rather not install anything, or you like a record you can search later. Skip it if a few minutes matter to you.

X

@codex_reset posts each alert (⚠️ ⌛️ ✅ 🎟️). It's also the one channel that posts Tibo's teases (👀), so it's a little busier than the others. The account checks the site's public data once a minute, so expect a short lag behind the instant channels.

Following the account is not enough to get notified. Turn on notifications for it in X, or the posts just wait in your timeline.

Pick X if you're on it anyway, or if you want the teases as well as the alerts.

Browser push

On codex-reset.com, open the alerts menu and turn on browser notifications. You don't need an account.

Push wakes your browser for the latest regular reset, strong Watch or banked update. If several things happen in a row, you may only see the latest state rather than every step. Old alerts never replay.

On iPhone, push only works after you add the site to your Home Screen and open it from there.

Push fits if you sit at the same computer all day and don't want another app.

Quieter options

The Atom feed. codex-reset.com/feed.xml lists Tibo's reset announcements. Each entry quotes his post and links to it on X. It's meant for feed readers and scripts. It records what he announced, not what our accounts measured, so for ✅ and 🔬 use one of the channels above.

The Tibo streams. If you want to read Tibo's posts yourself without opening X, Codex Reset · Tibo Radar on Telegram and #tibo on our Discord pass them along with little noise. They're separate from #alerts and don't mirror it. Think of them as context, not alerts.

Chinese readers can join our QQ group from the Chinese version of the site. It isn't an automatic alert channel, so pair it with one of the channels above.

Build your own alert

If none of these fit, the developer page documents a public JSON API and a read-only MCP server. In the page's words: "No key, no sign-up, no rate plan — the one condition is credit." If you show the data to other people, name the source and link it on the same surface ("Data: codex-reset.com"). A private script or a cron job that pings your own phone only needs a User-Agent that names your project.

The page's own example reads the current forecast:

curl -sS https://codex-reset.com/api/forecast \
  -A 'myproject/1.0 (+https://myproject.example)' \
  | jq '{p24: .probabilities.rounded_24h, p48: .probabilities.rounded_48h, last: .last_reset_at}'

If you'd rather ask your coding agent, add the MCP server. In Claude Code:

claude mcp add --transport http codex-reset https://codex-reset.com/mcp

Codex CLI and the IDE extension read ~/.codex/config.toml:

[mcp_servers.codex-reset]
url = "https://codex-reset.com/mcp"

Then ask "did Codex reset today?" between tasks. We walked through the terminal side in more detail in an earlier DEV post. The full reset history is also published as an open dataset under CC BY 4.0 on Hugging Face and Kaggle.

Which one to pick

• You want to know within minutes and Telegram is always open: Telegram.
• Your team is on Discord: Discord #alerts.
• You're at one desk all day and don't want to install anything: browser push.
• You want a searchable record and minutes don't matter: email.
• You're on X a lot, or you want the teases too: @codex_reset.
• You want it inside your own tooling: the API or MCP server.

One channel is enough. The main channels carry the same alerts, so subscribing to three just means hearing about the same reset three times.

What the alerts can't tell you

A few limits apply to every channel.

We monitor Pro and Plus. Our accounts are ChatGPT Pro and Plus. We don't cover Team or Business workspaces yet, so a ✅ tells you nothing about them.

A ✅ can rest on Tibo's word. If our monitors can't take a reading, a ✅ based on Tibo's confirmation says so in the message: "Our monitor hasn't measured it yet."

A Watch is a probability. 93% is likely, not certain. It's there to help you decide whether to wait or wrap up.

Some lanes trade detail for reliability. Email can very rarely miss a message, and push may show only the latest state. If you need every step, Telegram or Discord is the better fit.

Pick the channel you already check, and get back to work.

Codex Reset is an independent community tracker and is not affiliated with OpenAI. "Codex" is a trademark of OpenAI.