Codex Quietly Passed 15M Users, Got a Surprise Reset
On August 12, OpenAI’s Codex team lead Thibault Sottiaux — known in the community by his nickname Tibo — posted something cryptic on X: “I previously promised a reset for every 1M in additional active users for Codex, until 10M. We blew past that and have been silent since 10M. Little surprise for you tomorrow.” The next day, August 13, he revealed the punchline: “Old news actually from a bunch of days ago, but crossed that 15M. Enjoy a nice reset everyone. Landing in the next hour or so, go /fast.” The only independent corroboration currently available for these two posts comes from the tech blog explainx.ai and a thread on OpenAI’s own developer community forum, both of which quote identical wording — but neither links to the original posts. That makes 15 million, for now, a company-reported figure with no third-party audit behind it. Even so, the two posts pushed Codex’s reset saga — already running for most of the year — into a new, more dramatic chapter.
From “surprise tomorrow” to a make-good reset: the 15M milestone revealed in a day
Read together, Sottiaux’s two posts amount to an admission: the original “reset every million users” system quietly stopped producing public milestone announcements once Codex crossed 10 million on July 21. It took the vague “little surprise for you tomorrow” on August 12 for anyone outside OpenAI to learn something was even in the works. When the reveal came on August 13, the answer was that user growth had already blown past the old target — Codex had crossed 15 million users, and by Sottiaux’s own admission that number was “old news… from a bunch of days ago,” meaning it had already been true for a while before anyone said so publicly. Whether that 15 million figure counts Codex alone or folds in ChatGPT Work too wasn’t specified, and there’s no independent data to verify it either way. Either way, this reset arrived differently than the earlier round-number announcements — instead of firing the moment a milestone was hit, OpenAI teased it a day ahead and only delivered the actual reset afterward, which reads less like a scheduled perk and more like a deliberate bit of hype-building.
Not the first time: the reset race from 3 million to 10 million
Before this August surprise, Codex had already run through a long reset history on its way from 3 million to 10 million users. It started on April 7, when Sam Altman posted a tweet with almost no preamble: “To celebrate 3 million weekly Codex users, we are resetting usage limits. We will do this every million users up to 10 million. Happy building!” The promise wasn’t just talk. Codex had gone from 2 million to 3 million users in under a month, and Altman followed through with the first reset. Less than two weeks later, Codex crossed 4 million active users, and Altman posted again that same day: “Codex hit 4M active users, less than two weeks after hitting 3M. We will reset rate limits today!” According to follow-up posts from OpenAI team members on X and subsequent tech-media coverage, the pace barely let up over the following months — Codex passed 6 million and then 8 million users through June and July, before crossing the 10 million mark on July 21. OpenAI rolled out a fresh reset for paid Codex and ChatGPT Work users within an hour of announcing that milestone. From the first reset at 3 million to what was then the finish line at 10 million, OpenAI kept its word at every single round-number checkpoint — until public milestone announcements stopped, setting up August’s make-good reset.
June to July: the fastest growth, and the messiest stretch
Things got genuinely complicated after the July 9 launch of OpenAI’s GPT-5.6 model family — the flagship Sol, alongside the cheaper Terra and Luna tiers — plus the new ChatGPT Work product, which runs on Codex and GPT-5.6 together. Within days, users were complaining that usage was burning through much faster than expected, and a simultaneous desktop app redesign made things harder to use, not easier. Sottiaux stepped in on July 13 with three fixes at once: temporarily removing the “every 5 hours” short-term cap on Plus, Business, and Pro plans; tuning GPT-5.6 Sol’s inference to free up roughly 10% more usable quota; and issuing a straight reset to 6 million active users. About three hours later, Sottiaux posted again, admitting that some users who tried to redeem a banked reset during a two-hour window hadn’t received it, and since the system couldn’t precisely identify which accounts were affected, OpenAI issued another reset to everyone (reportedly around 500,000 ChatGPT Work and Codex users) who’d pressed the button during that stretch. Compared with the tidy plan to reset at each million-user milestone, this stretch looked a lot more like triage done on the fly.
How often does Codex actually reset? Even the trackers can’t agree
Going by the public record, Codex went through multiple resets and bulk resets between June and August 2026 — but “reset” itself covers several genuinely different kinds of events: official milestone celebrations, bug make-goods, batched banked-reset payouts, and promotional or post-incident credits. Lump them all together as one category and the count gets distorted fast. Several independent tracker sites have tried to tally a running total anyway, and they don’t agree with each other: one count puts Codex at 23 resets in 109 days, while a rival tracker, codex-reset.com, counts from September 2025 and arrives at 37 — a gap that comes down to different starting points and different counting rules, not one tracker being more “correct” than the other. What does seem clear, under any counting method, is that resets have been coming far more often than the round-number milestone announcements alone would suggest — and that’s before counting August’s make-good reset. That pace got frequent enough that OpenAI rolled out a “banked reset” feature: eligible users can save an earned reset instead of losing it, and apply it manually whenever they choose, rather than watching an unused weekly reset simply expire. According to OpenAI’s own developer community forum, banked resets obtained through things like referral promotions typically need to be used within 30 days of being credited, though the exact window depends on each promotion’s terms. The feature later expanded from the web app to mobile, turning what started as a one-off perk into something closer to a managed quota system.
Two very different developer reactions: Christmas, or free slot-machine spins?
In a Hacker News thread discussing the reset mechanism, developer reactions split cleanly into two camps. Some were simply grateful — one commenter described the feeling as “like Christmas,” and others said they finally felt free to use premium models like Sol that they’d normally ration carefully. Others were far more wary. One wrote: “I’ve stopped rationing and started spinning up a ton of agent runs… now I’m worried that once the resets stop, my normal workflow will blow right through the limit.” Some commenters were blunter still, comparing the string of resets to “free spins on a slot machine” — an incentive mechanism, not a gift — with a few speculating this ties into OpenAI’s potential IPO plans, though that particular claim remains community speculation with no confirmation from insiders or credible reporting. August’s tease-then-deliver playbook is only going to make that speculation harder to shake.
The backdrop: is there a price war with Chinese models in the mix?
Some developers in the community have also noted that this run of generous resets lined up roughly with an aggressive round of low pricing from Chinese model makers like Kimi K3 and DeepSeek. A few commenters speculated OpenAI might be reacting to that competitive pressure, swapping a direct price cut for “unlimited-feeling” usage to keep developers loyal to Codex. That said, this remains community observation and speculation — OpenAI hasn’t publicly tied its reset strategy to any specific competitor’s pricing, and the overlap could just as easily be timing coincidence or a parallel response to the same industry-wide pricing trend.
Codex’s reset run from 3 million all the way to 15 million users has genuinely given plenty of subscription-plan developers more headroom to run tasks without rationing. But once you stack the banked-reset system, the crisis-driven make-good resets, the milestone celebrations, and now an August surprise that was teased a day before it was delivered, what developers are getting looks less like a simple gift and more like a carefully engineered mechanism for locking in usage habits — how much of that is also about fending off competition remains, for now, observation and speculation rather than confirmed fact. The open question is what happens once growth slows and there’s no next milestone — or “little surprise” — left to tease: will developers who’ve gotten used to the extra headroom be willing to go back to their original limits?
