| --- |
| license: cc-by-4.0 |
| language: |
| - en |
| pretty_name: Codex Reset History |
| size_categories: |
| - n<1K |
| tags: |
| - codex |
| - openai |
| - software |
| - event-history |
| - usage-limits |
| - time-series |
| configs: |
| - config_name: default |
| data_files: |
| - split: train |
| path: codex-resets.csv |
| dataset_info: |
| features: |
| - name: event_id |
| dtype: string |
| - name: type |
| dtype: string |
| - name: announced_at |
| dtype: string |
| - name: date |
| dtype: string |
| - name: source |
| dtype: string |
| - name: post_url |
| dtype: string |
| - name: banked_state |
| dtype: string |
| - name: summary |
| dtype: string |
| --- |
| |
| # Codex Reset History |
|
|
| An independent, event-level dataset of Codex usage-limit reset announcements and banked reset grants since September 17, 2025. |
|
|
| **Canonical dataset, methodology and live updates: [codex-reset.com/timeline](https://codex-reset.com/timeline).** |
|
|
| This repository is a dated snapshot retrieved on **October 12, 2026 (Asia/Shanghai)**, not an automatically refreshed mirror. It contains **62 events: 52 resets and 10 banked grants**, with announcement dates from September 17, 2025 through October 11, 2026. For current counts and later corrections, use the canonical source. |
|
|
| ## Files |
|
|
| | File | Contents | |
| | --- | --- | |
| | `codex-resets.csv` | The event table: 62 rows and 8 columns; UTF-8 CSV. | |
| | `codex-resets.json` | The same events plus metadata, counts, provenance, and download links. | |
| | `datapackage.json` | The publisher's Frictionless Data Package schema; its resource URL points to the live canonical CSV. | |
|
|
| The `train` split used by the Hugging Face viewer contains the full archive. It is a packaging convention, not a benchmark train/test split. |
|
|
| ## Fields |
|
|
| | Field | Meaning | |
| | --- | --- | |
| | `event_id` | Stable string ID: an X post ID, an ID with `-banked` for an additional grant, or a publisher observation ID. Preserve this field as a string. | |
| | `type` | `reset` or `banked`. | |
| | `announced_at` | ISO 8601 UTC announcement timestamp. An observed-only event uses its recorded observation time. | |
| | `date` | UTC calendar date of `announced_at`. | |
| | `source` | `post` for a public source post, or `observed` for publisher monitoring without a source post. | |
| | `post_url` | Public source-post URL; empty for an observed-only grant. | |
| | `banked_state` | `announced`, `arriving`, `available`, or `unknown`; empty for a reset. | |
| | `summary` | English summary of the event. | |
|
|
| ## Sources and inclusion rules |
|
|
| The publisher curates public posts by [Thibault Sottiaux (@thsottiaux)](https://x.com/thsottiaux), with explicitly labeled observations from its monitoring accounts when no public post exists. This snapshot contains 61 source-post rows and one observed-only row. |
|
|
| The archive distinguishes usage-limit resets from banked credits for later use. Each banked grant is counted once. A post announcing both a reset and a banked grant contributes two events. Contextual notes, general promises, individual-account replies, and partial fixes are not intended to count as completed global resets. Banked grant states retain whether a grant was announced, arriving, or available. |
|
|
| According to the publisher's methodology, inclusion combines its alert judgement with manual review. Post timestamps derive from post IDs; monitoring accounts provide additional confirmation where available. Announcement time is not proof of the exact time every account received an event. |
|
|
| The canonical page records an October 10, 2026 reconciliation against other public trackers and explains its corrections. Historical inclusion decisions and completeness remain open to correction. |
|
|
| ## Suggested uses and limitations |
|
|
| Use this small curated archive for exploratory analysis of announcement timing, intervals between resets, banked-grant history, and AI software usage-limit policies. |
|
|
| It is not official OpenAI telemetry, individual usage data, a representative sample of all accounts, or a training/evaluation benchmark. It does not establish that every account received an event at the announcement time. Observed-only provenance is weaker than a public source post. The canonical archive can change as events and corrections are incorporated; this snapshot may become outdated. A pattern in past announcements does not guarantee future resets. |
|
|
| ## Example |
|
|
| ```python |
| import pandas as pd |
| |
| events = pd.read_csv("codex-resets.csv", dtype={"event_id": "string"}) |
| events["announced_at"] = pd.to_datetime(events["announced_at"], utc=True) |
| resets = events.loc[events["type"].eq("reset")].sort_values("announced_at") |
| gaps_days = resets["announced_at"].diff().dt.total_seconds() / 86400 |
| ``` |
|
|
| ## License and citation |
|
|
| Released by Codex Reset under **[CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)**. Provide visible attribution: **Data: [codex-reset.com](https://codex-reset.com/)**. Preserve source links and indicate any changes when reusing the data. |
|
|
| Suggested citation: Codex Reset (2026). *Codex reset history* [dataset]. https://codex-reset.com/timeline |
|
|
| Codex Reset is an independent community project and is not affiliated with OpenAI. “Codex” is a trademark of OpenAI. |
|
|
| ## Live source downloads |
|
|
| - [CSV](https://codex-reset.com/data/codex-resets.csv) |
| - [JSON](https://codex-reset.com/data/codex-resets.json) |
| - [Data Package](https://codex-reset.com/data/datapackage.json) |
| - [Timeline API](https://codex-reset.com/api/timeline) — includes contextual timeline material beyond the event-only export. |
|
|
| Questions or corrections: hello@codex-reset.com, or use the contact link on the canonical site. |
|
|
