Who runs this siteWHO
Codex Reset Radar is an independent developer project offering public reset signals, a personal-time helper and a downloadable event archive. It is not official or affiliated with OpenAI; personal allowance and exact reset times come from the account Usage Dashboard.
Content is produced two ways, labeled on every page: automated compilation — event data snapshotted from the codex-resets.com public archive (times, types, announcement text); and editorial — era context, guides and FAQs written by an editor from the facts in the announcements, re-reviewed on every data refresh.
Data & methodHOW
- Origin of every message: public posts on X by OpenAI Codex team members (Tibo / @thsottiaux et al.). The verdict does not depend on any single aggregation channel.
- Multi-source fusion: the homepage fuses X-direct original text, the codex-resets.com curated stream (attribution & thanks), news and community echoes. Only official channels can raise a status to OFFICIAL; old announcements return to archive context rather than generating a date forecast.
- Three advice levels: official announcements can activate SEND IT for the scope they explicitly name; otherwise the gauge only describes how the current gap compares with the 5.9-day historical average. A long gap is not an official probability or date guarantee. Plan scope is extracted from announcement text; when unstated it remains unknown rather than defaulting to all paid plans.
- Event archive: periodically snapshotted from the codex-resets.com free public API (currently 70 events, averaging ~5.9 days apart). The archive's raw sources are the same public announcements by OpenAI Codex team members on X.
- Daily digest: collected automatically from four free sources — Hacker News (Algolia API), Reddit public JSON, GitHub Releases, and OpenAI's official RSS — filtered by keywords and a minimum-heat threshold. No automated rewriting: items show their original titles and link to the original sources.
- Analysis integrity: per-event metrics (running average, gap percentile) use only the archive that existed before each event — no future data — so the pages show what was knowable at the time.
Why it existsWHY
Public reset announcements have no fixed schedule, while personal accounts carry their own limit times. This site separates them: sourced public signals, a local helper based on the Usage time you enter, and an open archive that publishes its data and baseline error instead of dressing a mean up as a forecast.
The erasPERIODS
| Era | Range | Character |
|---|---|---|
| Launch era | 2025-09-01 ~ 2025-10-31 | The archive begins on September 17, 2025 with a public reset event near the GPT-5-Codex la… |
| Winter compensation era | 2025-11-01 ~ 2026-01-31 | Announcements in this period cited a November outage recovery, usage discrepancies after a… |
| March incident cluster | 2026-02-01 ~ 2026-03-20 | Early March 2026 is an incident-dense segment of the archive: several events within nine d… |
| Growth era | 2026-03-21 ~ 2026-05-10 | Announcements in this period paired reset events with 3M weekly users, the $100 plan launc… |
| Spring tuning era | 2026-05-11 ~ 2026-07-08 | This segment records an accounting bug that drained limits, several small incidents and an… |
| July reset spree | 2026-07-09 ~ 2026-07-31 | This is the archive's densest three-week segment: more than ten public records, a temporar… |
| Sol era | 2026-08-01 ~ 2026-08-18 | Announcements around GPT-5.6 Sol included celebratory contexts such as an efficiency week.… |
| Milestones & banked credits | 2026-08-19 ~ 2026-09-02 | This segment records the 20M and 25M user milestones, an August 21 banked-reset grant and … |
| Astra era | 2026-09-03 ~ present | Astra-era announcements include banked resets tied to days waiting for access, an extra gr… |
CorrectionsCONTACT
The correction mailbox is awaiting forwarding setup. Until it is operational, this site does not publish a dead address or promise an unmonitored response time. Every event page keeps a clickable original source so readers can verify dates, scope and wording directly.