FULL STORY

Codex's 28-Day Improvement Sprint Kicks Off

Codex lead Thibault Sottiaux vowed daily improvements for 28 days or a full reset. Day two delivered: Auto-review went free for all ChatGPT users without quota usage.

2026-10-05 ~ 2026-10-07 · 2 episodes · 14 posts

Episode 1 · OpenAI Codex lead vows daily improvements for 28 days or full quota reset (2026-10-05, 8 posts)

On around October 5, OpenAI Codex lead Thibault (Tibo) Sottiaux made a rare high-stakes pledge on X: for the next 28 days, the team will either ship a significant improvement (not mere fixes) for the vast majority of Codex/Work users every day, or perform a full usage reset. This publicly binds product iteration pace to user compensation and is widely read as a bet on community trust.

Confirmed

  • The pledge, posted by Sottiaux on X around October 5, is corroborated by multiple posts (m2, m4, m5, m6, m7, m8).
  • Terms: daily significant improvements for Codex/Work users for 28 days, otherwise a full quota reset. aziz4ai's retelling says the reset would come on the final day; some users speculate a day without new features in late October would trigger it.
  • Directions: after soliciting community feedback, he named four priorities—simplifying the product, improving efficiency, breakthrough features, and new models. APPSO speculates the new model could be GPT-6.1 Astra.
  • Poster bdsqlsz issued a correction: OpenAI will grant 28 Codex quota resets over the 28 days; with usage halved, this amounts to roughly 14x the original usable quota.

Unconfirmed

  • The exact reset mechanics (daily trigger, single end-of-period reset, or one-off) vary across posts without official clarification.
  • The GPT-6.1 Astra link remains media speculation; bdsqlsz's 28-resets claim is a personal retelling pending official confirmation.

Why it matters

  • Large AI products rarely bind iteration cadence to user compensation so explicitly, putting the team under clear accountability.
  • Commenter kimmonismus sees the move as an attempt to stop user churn to Claude; APPSO notes growing skepticism, meaning delivery on the pledge will directly shape Codex's reputation and retention.

Episode 2 · OpenAI Makes Codex Auto-review Free for All ChatGPT Users (2026-10-06, 6 posts)

On Day 2 of its "28 Days of Improvements" launch season, OpenAI announced that Codex's Auto-review feature is now free for all users logged in with a ChatGPT account and no longer counts against plan usage quotas. The announcement was made on October 6 by Codex lead Thibault Sottiaux.

Confirmed

  • Auto-review is Codex's "Approve for me" permission mode: a separate second agent automatically reviews high-risk actions and makes approval decisions on the user's behalf; it can be enabled under Settings > Permissions
  • The feature is free for all ChatGPT-logged-in users and no longer consumes plan usage; previously, using this safety feature ate into users' quotas
  • @reachvb (an OpenAI employee) cleared up common community misconceptions about the feature and confirmed it is free
  • According to @thsottiaux's summary, Approve for me accounted for roughly 2-10% of usage when billed, so making it free is a better deal for users; Day 2 also brought updates including the Decisions API

Why it matters

  • Automated review is a safety feature, and previously consuming quota discouraged users from enabling it; making it free lowers the safety barrier and lets users save quota for core coding tasks
  • The update is part of OpenAI's intense 28-day release cadence, showing accelerating product iteration. While the community welcomed it, some complained that quotas weren't reset during the 28-day improvement streak