Codex raises weekly and hourly limits to 100% after 5 million users

• OpenAI reset Codex weekly and hourly quotas back to 100% across paid ChatGPT subscriptions, according to thsottiaux's reset announcement, with reach_vb's follow-up saying the reset applied across plans and surfaces.
• The reset landed one day after thsottiaux's dashboard teaser and alongside Sam Altman's repost of thsottiaux's 5 million users claim, which framed the quota change as a celebration of that milestone.
• Hands-on posts from theo's migration PR write-up, steipete's codemod note, and steipete's QA loop demo show where those quotas are going: large refactors, ad hoc codemods, and background test runs that keep opening fix PRs.
• The current Codex surface is wider than the quota-reset tweets imply. OpenAI's ChatGPT release notes, Business release notes, and Enterprise and Edu release notes already describe Windows computer use, mobile remote control, and usage profiles.
• Visibility still looks uneven. testingcatalog's profile-tab screenshot shows per-profile token stats, but an open GitHub issue on missing review-limit meters says some review quotas still are not exposed in the dashboard.

You can read the official release notes for the consumer app, compare them with the Business changelog, and then jump straight to the weird use cases: steipete has Codex running background QA through webVNC and browser-use tools, nummanali is using remote desktop sessions for family IT support, and reach_vb got it to stitch MP4s by checking codec, fps, and resolution first.

link

Tibo Sottiaux said paid ChatGPT subscribers were back to 100% weekly and 100% hourly limits. reach_vb's post added that the reset covered multiple Codex surfaces, not just one app view.

The timing was explicit. thsottiaux's May 30 post teased a dashboard number, then Sam Altman's repost repeated the 5 million user line and the plan to reset limits the next morning.

That 5 million figure is a user count, not a usage chart. petergostev's projection chart extrapolated to 10 million by October, while steipete's repost of _simonsmith noted that 5 million would still be a small slice of ChatGPT's overall base.

The fastest signal in the evidence pool is not "more messages," it is bigger jobs. Theo said a migration PR that previously would have taken a week and multiple reviewers got knocked out in a few hours, then followed up that an 18,000 line move had merged with no regressions so far.

The pattern shows up outside one post. steipete's codemod note said Codex wrote ad hoc codemods for a larger TypeScript migration, which is a more specific claim than generic "helped with refactoring" praise.

Codex is also getting used as a long-running worker, not just a chat window. Steipete said he had trained it to generate a user-test scenario for every commit, run those checks through webVNC and browser-use tooling, and open PRs with fixes in the background.

That lines up with the way OpenAI has been widening the product surface in its ChatGPT release notes, which describe remote control, mobile steering, and computer-use features that make continuous runs more plausible than short prompt-response sessions.

A related repost from mattlam_'s repost of Jinjing Liang framed the best use of the reset as overnight scanning for bugs or security issues. That is still anecdotal, but it matches the quota pressure you would expect from background review loops.

OpenAI's public changelogs say Codex now supports Windows computer use and remote control from ChatGPT on iOS or Android, or from Codex on Mac, across the main ChatGPT release notes, the Business release notes, and the Enterprise and Edu release notes. nummanali's post is the clearest user-level example in the evidence pool: logging into a brother's Codex session on another laptop and handling support remotely.

The rough edges are already visible in the repo issues:

• A Linux SSH remote-control issue says ChatGPT mobile could not connect to a physical Linux remote host on newer Codex versions even while Codex Desktop still worked.
• A WSL availability issue says Windows Desktop could do Remote Control with the app server in WSL, but Computer Use and Browser Use stayed unavailable with a wsl-disabled reason.

reach_vb's MP4 stitching example is a useful oddball here because it shows the same "remote worker" behavior on a smaller task: find the files, verify codec compatibility, concatenate without re-encoding, then decode and spot-check the result.

One quiet product detail matters more after a quota reset: metering. testingcatalog spotted a new Profile tab with detailed token-consumption stats, and OpenAI's ChatGPT release notes say usage profiles are part of the latest Codex update.

The visibility story is incomplete. An open GitHub issue on missing review-limit meters says users can still hit a Codex review limit without seeing that quota in the dashboard, even though the expected behavior was for review usage limits to be visible.

That split, more generous quotas plus better token stats, but still patchy limit exposure, is a better read on the moment than the celebration posts alone.