--- title: "Codex Usage Limit Resets: Timeline and How to Track the Next" description: "A sourced timeline of the two 2026 Codex usage limit resets, how short-window limits differ from weekly limits, and how to track the next reset with cdnping." url: https://cdnping.org/en/blog/codex-usage-limit-reset-timeline language: en published: 2026-10-17 category: "Service updates" tags: ["Codex","OpenAI","usage limits","limit reset","service monitoring"] source: cdnping.org license: Free to quote and summarize with a link to the url above. --- # Codex Usage Limit Resets: Timeline and How to Track the Next As of October 17, 2026, the cdnping timeline lists only two Codex usage limit resets: April 8, 2026 and July 29, 2026. Both were ad hoc, all-user resets announced by OpenAI, with no fixed schedule. To hear about the next one early, subscribe to the RSS feed on cdnping [/changes](/en/changes) and watch the user reports on [/down/codex](/en/down/codex). ## How many times has Codex reset usage limits? The timeline only includes events that meet two conditions: there is public reporting or an official statement, and the reset applied to all users. Individual posts on social media along the lines of "my limit seems to have refreshed" don't count. | Time (UTC) | Beijing time | Event | Source | |---|---|---|---| | 2026-04-08 | 2026-04-08 | Sam Altman announced that Codex had reached 3 million weekly active users and reset usage limits for all users, promising another reset for every additional 1 million users, up to 10 million | Business Today report | | 2026-07-29 04:33 | 2026-07-29 12:33 | The head of Codex, Tibo, announced that usage limits had been reset for all ChatGPT Work and Codex users | Relayed by a KuCoin news flash | The first report is only precise to the day, so the table gives no time of day. Both entries come from secondhand media coverage; treat OpenAI's official channels as the authoritative statement. ### What does the April reset have to do with user milestones? The April 8 reset was tied to Codex reaching 3 million weekly active users. The announcement came with a promise: another reset for every additional 1 million weekly actives, up to 10 million. Read literally, more resets may follow, but their timing depends on when the user count crosses the next million, which can't be predicted from outside. We found no reliable source for resets at any of the other milestones, which is why the timeline has only this one. If you see an article listing resets at 4 million or 5 million, find the original source before trusting it. ### What was different about the July reset? The July 29 reset was announced by the head of Codex. It was explicitly scoped to ChatGPT Work and Codex users and mentioned no user milestone. So resets are not necessarily tied to milestones. ### Can you predict the next one from these two? - The two resets were about three and a half months apart, but two data points are not enough to project the next date. - In both cases a lead publicly announced the reset first and the media relayed it afterward. Official channels and changelogs are the earliest place to see the news. - Neither report says whether the underlying accounting period shifted after the reset. Next time, note what your usage display shows before and after. ## How do you tell a short-window recovery from a weekly limit reset? Many "my limit was reset" discussions mix up two different kinds of limits. What follows is only the general concept. Actual durations and numbers vary by plan and over time, so rely on OpenAI's official documentation and what your account shows. **Short-window limit** (the community's "5-hour limit" is of this kind): caps usage within a short period and recovers on its own when the window expires. If you get throttled after several hours of continuous use and can work again after a wait, that is window recovery, not a reset. **Weekly limit**: covers a longer period. Even after the short window recovers, you will still be restricted if the weekly allowance is used up, until that period ends. When an all-user reset happens, this is typically the part that gets restored early, though the official announcement is the authority. ### Cheat sheet: which case am I seeing? | What you see | Likely cause | What to do | |---|---|---| | You hit a limit after several hours of use, and it recovers on its own after a while | Short window expired and recovered | Nothing to do; note the recovery time | | It recovers briefly, then you are limited again soon, with a long wait time | Weekly limit nearly used up | Check the weekly limit shown in your account and pace your usage | | The weekly limit suddenly refills before the period ends | Possibly an all-user reset | Check /changes and official channels, and see on /down/codex whether others noticed at the same time | | Your limit hasn't changed, but many requests fail or are slow | Service outage, unrelated to limits | Check /status/openai and OpenAI's official status page | The last row is the easiest to get wrong. A rate-limit error and a service outage can look similar from the client, as the status page records in the next section show. ## Which recent OpenAI incidents are easy to mistake for limit problems? cdnping syncs OpenAI's official status page every 15 minutes and shows the results at [/status/openai](/en/status/openai). Per the synced data, since September 17, 2026 the OpenAI status page has recorded 4 major, 19 minor, and 2 no-impact incidents. Several from the past week that involved Codex or had broad impact are below (source: OpenAI's official status page, synced by cdnping; durations are as recorded on the status page): | Start (UTC) | Incident title | Severity | Duration | |---|---|---|---| | 2026-10-15 16:29 | Widespread feature degradation | major | about 146 minutes | | 2026-10-14 07:05 | Degraded performance in Codex and ChatGPT Work | minor | about 167 minutes | | 2026-10-13 08:10 | Degraded performance for a model in Codex and Work modes | minor | about 506 minutes | The October 13 incident lasted more than 8 hours. If your Codex tasks kept slowing down or getting interrupted during that time, service degradation is a more likely cause than your usage limit. Check the status page first, then think about limits. On API endpoint connectivity: in the window from 08:40 to 10:05 UTC on October 17, 2026, cdnping probed `api.openai.com` from the Cloudflare Denver node (DEN). All probes succeeded, with a median latency of 96 ms and a P95 of 194 ms. cdnping only started collecting that day, so this shows only that the endpoint was reachable during that hour and a half. It says nothing about long-term stability. ## How can you find out about the next reset right away? The steps below use only public pages, and cdnping itself needs no login. ### Step 1: Subscribe to the /changes RSS feed [/changes](/en/changes) monitors the official changelogs, pricing pages, and news pages of 11 sources, including DeepSeek, OpenAI, Codex, Claude, and Gemini, and records additions and deletions line by line. If the Codex changelog adds an entry about usage limits, it will show up here. 1. Open [/changes](/en/changes), find the RSS feed URL on the page, and copy it. 2. Add it to your RSS reader. 3. If your reader supports keyword filters, set `Codex`, `usage`, `limit`, `reset`. If you live in the terminal, you can pull the RSS once with curl to confirm it works, then hand it to a scheduled job: ```bash # Replace with the actual URL given on the /changes page curl -s "" | grep -iE "codex|usage|limit|reset" | head -20 ``` ### Step 2: Bookmark /down/codex and contribute reports [/down/codex](/en/down/codex) combines the official status page, an API probe every 5 minutes, and one-click user reports. The report types include "limit reset", as well as "rate limited", "slow", "logged out", and others. - When your limit suddenly refills, click "limit reset". If many people report it within a short time, it is most likely an all-user reset rather than your own period rolling over. - When you are rate limited, first see whether others on the page report the same problem, then decide whether to wait or check your own configuration. ### Step 3: Check /status/openai when requests fail [/status/openai](/en/status/openai) shows edge probe results for the official API endpoint and status page incidents from the past 30 days. If your limit looks normal but requests keep failing, look here first. ### Step 4: Read the weekly stability report [/reports](/en/reports) is generated automatically every Monday and can be subscribed to. If you don't want to watch things daily, read it once each Monday to see whether last week had any Codex-related incidents. ### Self-check list When your usage behaves oddly, go through these in order: - [ ] Is the limit shown in your account the short-window limit or the weekly limit? - [ ] Are there any recent new entries for Codex or OpenAI on /changes? - [ ] On /down/codex, has anyone else reported "limit reset" or "rate limited" in the past hour? - [ ] Are there any ongoing incidents on /status/openai or OpenAI's official status page? - [ ] Is your local network fine? The live speed test on the cdnping home page ([/](/en)) shows the initial connection and round-trip time to the OpenAI API, so you can rule out your own network. Also, write down your current limit display and the time in a note. When the next reset happens, you can compare and see exactly what it changed. ## What if you keep running out of usage? If your workflow often stalls on limits, besides waiting for a reset you can consider other ways to access models. For example, [AGI Plan](https://agiplan.dev) (AGI Plan is a commercial partner of cdnping) offers a step-by-step Codex setup guide where changing a single endpoint address is enough, and no ChatGPT account is needed. It is a multi-provider LLM aggregation gateway, and states plainly that it is not an officially authorized channel. Plans are on the [pricing page](https://agiplan.dev/pricing), and available models are listed on the [models page](https://agiplan.dev/models). Whatever third-party service you use, follow its terms of service and local regulations.