Skip to content
Technology

OpenAI Promises Codex Reset as Claude Pro Users Hit Limits

OpenAI's Codex lead pledged a daily fix or reset for 28 days, as Anthropic's own increase to Claude Pro nets out as a cut.

Edited by Axel6 min read · 1,362 words · 12 sources
A digital dashboard fuel gauge glowing empty, its warning light lit
A digital dashboard fuel gauge glowing empty, its warning light lit. Photo: Deloreanman14 / Wikimedia Commons, CC BY-SA 4.0
“OpenAI's Codex team promised to ship a fix or a full reset every single day for 28 days straight. Six of the first thirteen days brought only the reset. Does a month of promised resets prove OpenAI fixed what was draining every user's quota?”

Thibault Sottiaux runs Codex and ChatGPT Work at OpenAI. On Oct. 5, he made a public bet. "Over the next 28 days, each day we'll either ship one thing that is a clear improvement and relevant for most codex/work users or ship a full reset," he wrote on X1. Day 13 of that pledge landed on Oct. 17. It brought only the reset. "Day 13/ Le limit reset," Sottiaux posted that afternoon2. Six of the pledge's first thirteen days have now ended the same way, a cleared quota standing in for a shipped fix5. Anthropic's Claude Pro subscribers are living a quieter version of the same story. Both companies sell premium access to AI coding help. Each spent 2026 publicly rewriting how much of it a paying user actually gets.

Thirteen Days, Six Resets

Sottiaux's pledge runs through roughly Nov. 1. Each day is supposed to bring a product improvement, a full reset, or both5. The independent tracker Codex Resets logs every announcement against that promise4. Its ledger for the first thirteen days shows nine with an improvement. Six brought a reset, two of those on days that also shipped an improvement5.

A vintage mechanical stopwatch, stopped mid-run
A mechanical stopwatch, the kind of clock that keeps Codex's five-hour session window and Claude Pro's rolling limits running, keyboard or idle.Ansgar Koreng / Wikimedia Commons, CC BY-SA 4.0

A reset clears whatever a Codex or ChatGPT Work subscriber had already burned through that week. It costs OpenAI little to promise in public. Users pay even less to collect it. Day 2 brought a reset and an improvement together. So did Day 3. Days 6, 8, 10 and 13 brought a reset alone5. Codex Resets counts 63 resets since it started tracking. The gaps run 6.4 days apart on average, with the longest stretch reaching 67.7 days4.

OpenAI ships a reset on 6 of the Codex pledge's first 13 days

Daily outcomes in OpenAI's 28-day Codex and ChatGPT Work pledge, Oct. 5 to Oct. 17, 2026, one cube per day

Each cube is one pledged day: magenta cubes are days that brought only a reset, the blue cubes mark the two days that brought a reset and an improvement together, and gray cubes are days that brought an improvement alone.

Data through Oct. 17, 2026.

Source: Codex Resets [5]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
Reset only4
Reset and improvement together2
Improvement only7
Whole13

That average matters, because a reset only helps someone who already hit a wall. A six-day gap means roughly six days of normal use before a quota needs rescuing again. The pledge turns an irregular, apologetic event into a scheduled one. It stays silent on whether Codex now gives a user more to work with than it did in September.

What a Reset Fails to Prove

Curtis Pyke runs Kingy AI. He spent August digging into whether OpenAI had quietly cut Codex's limits. His conclusion sat between the two loudest camps. "OpenAI now confirms that poorer cache performance may be causing faster effective depletion for some Codex users, but the evidence does not establish that OpenAI quietly halved Codex limits systemwide," he wrote6.

A parking meter stands on a Washington, D.C. street
A parking meter in Washington, D.C. counts down paid minutes, the same way a Codex or ChatGPT Work plan counts down a quota that a reset clears, leaving the cause unexplained.Kurt Kaiser / Wikimedia Commons, CC0

Pyke's reporting found real mechanisms behind the complaints. OpenAI had acknowledged worse cache hit rates for some accounts. It also pointed to a newer model, Sol, that burned through limits faster than the one before it6. What eluded Pyke was a company-wide cut, published anywhere, that matched the scale of the complaints. A promised reset answers the symptom. It leaves the mechanism, cache performance, model choice, raw compute, to whichever thread a user happens to read.

A car's odometer display showing its running mileage total
An odometer keeps a running count that only climbs. Codex Resets keeps the same kind of tally on OpenAI's pledge, 63 resets logged as of Oct. 17, 2026.William J Sisti from Morristown, NJ, USA / Wikimedia Commons, CC BY-SA 2.0

The OpenAI Developer Community's own Day 13 thread shows the same gap between relief and explanation. One member, posting under the name VeitB, summarized the pledge plainly: "Starting today, there will be one clear improvement each day for most of all Codex or Work users. On days without one, everyone gets a full reset instead and sometimes we get both"7. The thread's replies ran toward a specific complaint: a $200-a-month Codex Pro plan exhausted in two days. A reset clears that quota. It leaves the cause unexplained.

Claude Pro's Quieter Version of the Same Problem

Anthropic has run its own version of this argument since the summer of 2025. On July 28 that year, the company announced new weekly caps for Claude Code on its Pro and Max plans. It cited unprecedented demand8. Amie Rotherham, an Anthropic spokesperson, told TechCrunch at the time: "Claude Code has experienced unprecedented demand since launch. Most users won't notice a difference"8.

Two electricians work on a building's electricity meter
Electricians adjust a building's electricity meter. Anthropic made a comparable adjustment to Claude Code's weekly limits on Aug. 29, 2026, calling a net cut a 25 percent increase.Jeromi Mikhael / Wikimedia Commons, CC0

Nine months later, Anthropic reversed course. On May 6, 2026, it doubled Claude Code's five-hour rate limits for Pro, Max, Team and Enterprise plans. It also dropped a peak-hours reduction it had added for Pro and Max accounts9. That doubling lasted through the summer as a temporary boost. On Aug. 29, Anthropic made a smaller version permanent instead. It raised the standing limit by 25 percent and framed the change as an upgrade10. Compared with the boosted limits users already had, the company said plainly that the change "works out to a 17% reduction in weekly limits on Claude Code"10.

Jean Leon covers AI tools for Android Headlines. He read that announcement the way Pyke read OpenAI's resets: a company explaining a cut in the language of a gift. "Trying to spin a usage drop as a 25% upgrade just because you are ending a temporary promotion is corporate marketing at its worst," Leon wrote11. He added a line that reads as a direct answer to Sottiaux's pledge: "Anthropic's quotas are already so opaque that trying to budget your remaining weekly prompts feels like playing a guessing game"11.

OpenAI and Anthropic have changed coding limits 5 times since mid-2025

Dated changes to Claude Code and Codex usage limits, from each company's own posts, July 2025 to October 2026

Each dot is a dated change: magenta dots are the two that cut what a subscriber actually gets once the fine print is read, gray dots are the three that added capacity or promised a process for restoring it.

Data through Oct. 17, 2026.

Sources: Thibault Sottiaux on X [1]; Thibault Sottiaux on X [2]; Codex Resets [5]; Anthropic [9]; BleepingComputer [10]. Chart by The AILately.com Desk.

The numbers behind this chart
DateEvent
July 28, 2025Anthropic adds Claude Code caps. New weekly rate limits for Pro and Max, citing unprecedented demand.
May 6, 2026Anthropic doubles the 5-hour limit. Pro, Max, Team and Enterprise get double Claude Code capacity, and the peak-hours cut drops.
Aug. 29, 2026A 25% rise nets a 17% cut. Anthropic's own math: the new permanent limit sits 17 percent below the summer boost it replaced.
Oct. 5, 2026Sottiaux pledges 28 days. OpenAI's Codex lead promises a fix or a full reset every day through roughly Nov. 1.
Oct. 17, 2026Day 13 brings only a reset. The sixth reset day of the pledge's first thirteen, with the reset standing alone.

A GitHub issue against Anthropic's own Claude Code repository carries a blunt title: "Weekly Usage Limits Making Claude Subscriptions Unusable." It opened in October 202512. It has stayed open through round after round of the same complaint. A Pro or Max subscriber burns a week's quota in a day or two of normal use12. The issue predates Anthropic's May increase. It survived that change. The thread is still collecting comments after the August adjustment.

The Case For and the Case Against

Sottiaux's pledge makes OpenAI's case for him. A company that ships a visible reset on the days its fixes fall short is choosing an audit trail over silence. His own words set the terms before a single day passed: "Over the next 28 days, each day we'll either ship one thing that is a clear improvement and relevant for most codex/work users or ship a full reset. Let the improvements begin"1. Day 13 kept that promise, even on a day with only the reset to show for it2.

Leon's reporting makes the opposite case. It aims at the industry pattern, rather than one company. A pledge or a percentage increase announced in public still leaves the reader guessing at the number underneath it. "Cutting their working capacity by 17% right in the middle of active projects is going to burn a lot of goodwill," Leon wrote of Anthropic's August change. The sentence reads just as well pointed at OpenAI's own ledger of resets standing in for fixes11.

Read together, the two cases measure different things. Sottiaux answers for the process: OpenAI will tell you something happened, every day, for 28 days. Leon answers for the number. A told-you-something still has to add up to more access than less, or the telling becomes the whole product.

By the numbers

  • 28: the number of days Sottiaux pledged an improvement, a reset, or both, starting Oct. 5, 202615.
  • 6 of 13: the days into that pledge that had brought a reset as of Oct. 175.
  • 63: resets Codex Resets has logged in total, 6.4 days apart on average4.
  • 67.7 days: the longest gap the tracker has recorded between two resets4.
  • 17 percent: the net cut Anthropic itself attached to its "25% increase" to Claude Code's weekly limits on Aug. 29, 202610.
  • 2x: the five-hour rate limit increase Anthropic gave Claude Code's Pro and Max plans on May 6, 20269. A smaller, permanent gain replaced it in August10.
  • 15 months: the span from Anthropic's first Claude Code caps, announced July 28, 2025, to Sottiaux's Oct. 5, 2026 pledge81.

What to watch

Watch whether Codex Resets logs an improvement-only stretch of three days or more before Nov. 145. That run would say OpenAI's fixes are gaining ground on its complaints. The GitHub issue against Claude Code's weekly limits is worth checking too. Look for a comment from Anthropic itself, naming the hours or tokens a Pro subscriber can expect in an ordinary week12. That would be more specific than a percentage. A published number, from either company, would settle more than another reset.

Sources

  1. Thibault Sottiaux (@thsottiaux), "Over the next 28 days, each day we'll either ship one thing that is a clear improvement and relevant for most codex/work users or ship a full reset. Let the improvements begin.," X, Oct. 5, 2026, https://x.com/thsottiaux/status/2106845241357824205
  2. Thibault Sottiaux (@thsottiaux), "Day 13/ Le limit reset," X, Oct. 17, 2026, https://x.com/thsottiaux/status/2111492673890824348
  3. Thibault Sottiaux (@thsottiaux), "Landing sometime today by EOD PST," X, Oct. 17, 2026, https://x.com/thsottiaux/status/2111492738025959922
  4. Codex Resets, "Codex Limit Reset Tracker & History," Codex Resets, Oct. 17, 2026, https://codex-resets.com/
  5. Codex Resets, "Tibo's 28-day Codex challenge," Codex Resets, Oct. 17, 2026, https://codex-resets.com/tibo-28
  6. Curtis Pyke, "Were Codex Limits Reduced? What the Evidence Shows," Kingy AI, Aug. 20, 2026, https://kingy.ai/blog/codex-limits-reduced-investigation/
  7. VeitB, "Reset: Day 13 of 28 days of Quality of Life improvements and/or a full Reset," OpenAI Developer Community, Oct. 17, 2026, https://community.openai.com/t/reset-day-13-of-28-days-of-quality-of-life-improvements-and-or-a-full-reset/1403525
  8. Maxwell Zeff, "Anthropic unveils new rate limits to curb Claude Code power users," TechCrunch, Jul. 28, 2025, https://techcrunch.com/2025/07/28/anthropic-unveils-new-rate-limits-to-curb-claude-code-power-users
  9. Anthropic, "Higher usage limits and a SpaceX compute deal," Anthropic, May 6, 2026, https://www.anthropic.com/news/higher-limits-spacex
  10. Mayank Parmar, "Anthropic is cutting Claude Code's current weekly limits by 17%," BleepingComputer, Aug. 29, 2026, https://www.bleepingcomputer.com/news/artificial-intelligence/anthropic-is-cutting-claude-codes-current-weekly-limits-by-17-percent/
  11. Jean Leon, "Anthropic's Claude Code Usage Cap Update Actually Hides a Quiet Downgrade," Android Headlines, Aug. 30, 2026, https://www.androidheadlines.com/2026/08/anthropic-claude-code-weekly-limits-update.html
  12. mgarbs and contributors, "Weekly Usage Limits Making Claude Subscriptions Unusable," GitHub, Oct. 13, 2025, https://github.com/anthropics/claude-code/issues/9424

Cite this piece

Ryan Elliott Dennis, "OpenAI Promises Codex Reset as Claude Pro Users Hit Limits," AI Lately, Oct 17, 2026, https://ailately.com/articles/openai-codex-reset-claude-pro-limits

Tags: OpenAI · Codex · Anthropic · Claude Pro · Claude Code · usage limits · Thibault Sottiaux

Related, lately

More

Keyboard

j k
Move through a list
Enter
Open the selected piece
/
Search, and browse the columns
g then r
Right Now
g then a
Articles
g then s
The Signal
g then o
Opinion
g then b
Analysis
g then p
People
?
This sheet
Esc
Close