gpt-6.2-astra in Codex usage: which part is real
As of 2026-11-02 OpenAI has not released GPT-6.2: the official model page lists the Astra id as gpt-6-astra, the newest model entry in the API changelog is GPT-6 Rosalind on 2026-10-30, and the Responses API reference contains no gpt-6.2. Since 2026-11-01 the string has been showing up in Codex usage breakdowns and API responses posted by several users, while the model picker still reads GPT-6 Astra (@leo114119, @Mr_Salio, 2026-11-01); those are community claims, unconfirmed by OpenAI.
Updated 2026-11-02
One key for Claude, GPT and leading Chinese models, billed per token, up and running in minutes.
- Billed per token — live rates on /models
- Pay by card (Visa / Mastercard), Apple Pay, Google Pay or crypto
- Self-service: top up and activate instantly
Four points, each labelled with its source
No 6.2 in official pages
In the texts fetched on 2026-11-02 - the OpenAI API changelog, the Responses API reference, the Codex model docs and the 10 most recent releases in openai/codex - gpt-6.2 produces 0 hits.
The day the screenshots started
Community screenshots began on 2026-11-01: @leo114119 said the picker shows GPT-6 Astra while the usage data shows Gpt 6.2 Astra, and @Mr_Salio said the API response breakdown shows gpt-6.2-astra. Both are community posts, unconfirmed by OpenAI.
The default model in Codex release notes
Official release notes: rust-v0.161.0 (2026-10-07) made GPT-6.1 Sol the default model in the bundled and Amazon Bedrock catalogs, and the current latest stable rust-v0.166.0 (2026-10-29) only added GPT-6.1 Terra, and its notes do not mention 6.2.
The official way to see the active model
Codex docs state that /model chooses the active model, /status displays session configuration and token usage and confirms the active model, and codex debug models prints the catalog Codex sees - add --bundled to list only what ships with the binary.
What the string is, and what it is not
As of 2026-11-02 gpt-6.2-astra is not a model id OpenAI has released: the model values in the Responses API reference are gpt-6-astra, gpt-6.1-sol and gpt-6.1-terra - there is no 6.2. The string appears in Codex usage breakdowns and API responses posted since 2026-11-01 (@leo114119, @Mr_Salio, @Mashiro_0619, @AGTPinsights), unconfirmed by OpenAI. Why that is not release evidence: the same command gives opposite results - @ai_funss (2026-11-01) says codex --model gpt-6.2-astra runs, @PokeBenchTV replied the same day that you can put anything you want in there, @EthanAgentLab (2026-11-02) says some accounts get model not supported, and artificialwatch.com (2026-11-01) reports its own 400 model_not_found. cellcog.ai (2026-11-01) concludes that what exists is a model id in a usage chart, and a usage label alone cannot show which model ran. All of it is community content; OpenAI has published nothing.
2026-11-01: screenshots, not an announcement
The posts circulating since 2026-11-01 fall into three groups. Usage: @leo114119 and @Mashiro_0619 posted Codex usage charts, the latter saying the image came from three users; @Mr_Salio posted gpt-6.2-astra in an API response breakdown; @AGTPinsights said more accounts showed it over the weekend, reading it as a wider gray rollout. Command: @ai_funss posted codex --model gpt-6.2-astra running in a terminal. The fullest account is @EthanAgentLab (2026-11-02): the status bar read gpt-6.2-astra medium on Codex CLI v0.166.0, the same flag does not start on every account, some return model not supported, and the author's own reading is a quiet partial rollout rather than a launch. OpenAI has confirmed none of it, and the name has not appeared on an official page to date.
Timeline (each line names its source)
codex rust-v0.161.0 is released: official notes make GPT-6.1 Sol the default model in the bundled and Amazon Bedrock catalogs (official releases).
The current latest stable rust-v0.166.0 is released: the notes add GPT-6.1 Terra to the catalogs and say nothing about 6.2 (official releases).
Users start posting gpt-6.2-astra from Codex usage; rebuttals and model not supported reports come the same day, unconfirmed by OpenAI. This page will be updated once OpenAI ships.
Three columns: what is confirmed, what is claimed, what to do
Confirmed (official pages, verbatim)
1. The official model page writes the Astra id as gpt-6-astra. 2. The Responses API reference describes model as the model id used to generate the response, with gpt-6-astra, gpt-6.1-sol and gpt-6.1-terra as values. 3. The newest model entries in the API changelog are GPT-6.1 Terra on 2026-10-26 and GPT-6 Rosalind on 2026-10-30. 4. Codex docs require Codex CLI 0.159.0 or later for GPT-6.1 Terra, and the configuration reference says to set model to one available to your signed-in account or workspace. 5. GitHub releases show the current latest stable as rust-v0.166.0 (2026-10-29), whose notes add GPT-6.1 Terra to the catalogs and do not mention 6.2; the release notes naming GPT-6.1 Sol as the default model are those of rust-v0.161.0 (2026-10-07).
Unconfirmed (community claims, attributed)
All of the following is community content that OpenAI has not confirmed: @leo114119 (2026-11-01) wrote that Codex is reportedly in a gray rollout, and says the usage data shows Gpt 6.2 Astra; @Mashiro_0619 (2026-11-01) claims many users got it and posts charts from three users; @Mr_Salio (2026-11-01) reports gpt-6.2-astra in the response breakdown; @ai_funss (2026-11-01) says codex --model gpt-6.2-astra runs directly; @AGTPinsights (2026-11-01) reads it as a wider gray rollout; @EthanAgentLab (2026-11-02) says the same command returns model not supported on some accounts. The pushback is attributed too: @PokeBenchTV (2026-11-01) says you can put anything you want in there; artificialwatch.com (2026-11-01) reports its own 400 model_not_found; cellcog.ai (2026-11-01) writes that a usage label alone cannot show which model ran; thewincentral.com (2026-11-01) calls it a limited stealth rollout, which is inference. None of these has an official source.
Confirmable and not confirmable
gpt-6.2-astra in the usage chart
Confirmable: since 2026-11-01 people have seen this line in Codex usage breakdowns and API responses, and official pages do not list it. Not confirmable: whether a new set of weights is behind the name, how many accounts are covered, or when it ships - OpenAI has written none of that, and the dates and shares circulating have no source.
The ids to use in code today
The ids listed on the official model pages and in the Responses reference are gpt-6-astra, gpt-6-sol, gpt-6-luna, gpt-6.1-sol and gpt-6.1-terra. OpenAI's own wording for retiring an older model is the same pattern: replace it with a model available to your account and client, and update scripts and codex exec --model references.
How to check which model you are on, and what to do now
1. Read the model field in the API response: the reference states it is the model id used to generate the response. 2. In an interactive Codex session use /model to choose the active model; /status displays session configuration and token usage and lets you confirm it. 3. To see the catalog itself, run codex debug models for JSON output, and add --bundled to inspect only the catalog shipped with the binary. 4. A name in the usage breakdown is only a label: cellcog.ai (2026-11-01, community) concludes a usage label alone cannot show which model ran, and OpenAI has not confirmed anything. 5. In production, until OpenAI ships, keep the ids the official model pages list - gpt-6-astra, gpt-6.1-sol, gpt-6.1-terra - and do not repoint routing or add automatic switching for a string from a screenshot.
What id to write when Codex runs on QCode
Per the Codex page on docs.qcode.cc: GPT runs on the OpenAI Responses protocol, so config.toml sets base_url = https://api.qcode.cc/openai and wire_api = responses, and model takes the listed ids gpt-6.1-sol, gpt-6.1-terra, gpt-6-sol, gpt-6-astra and gpt-5.6-terra; QCode does not support gpt-6-luna or gpt-5.6-luna for now, so please use another GPT model. Billing is per token, with per-model rates on /models. As of 2026-11-02 the /models list has no GPT-6.2, and new listings follow /models. Not ready for a monthly plan? Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 days; if you already have a Personal Plan, add a booster pack when you need more quota.
Frequently asked questions
Is gpt-6.2-astra in my Codex usage a released model?
No. As of 2026-11-02 OpenAI has not released GPT-6.2: the model page writes gpt-6-astra for Astra, the newest model entry in the API changelog is GPT-6 Rosalind on 2026-10-30, and the Responses reference has no gpt-6.2. The name comes from screenshots users have been posting since 2026-11-01, unconfirmed by OpenAI.
Someone shows the command working. Is that release evidence?
No. The official description of the model setting is to set it to one available to your signed-in account or workspace, with gpt-6.1-sol as the example. Within the community, @PokeBenchTV (2026-11-01) says you can put anything you want in there, @EthanAgentLab (2026-11-02) says the same command returns model not supported on some accounts, and artificialwatch.com (2026-11-01) reports a 400 model_not_found of its own. Accepting a string is not the same as being able to call it, and calling it is not a release; OpenAI has confirmed none of this.
How do I check which model served my request?
Read the model field in the API response, which the reference defines as the model id used to generate the response. In Codex use /model to switch, and /status for session configuration and token usage; for the catalog itself run codex debug models, adding --bundled to limit it to the shipped catalog. A line in the usage breakdown is only a label.
Should I change code or replan for it?
No. Until it ships, keep the ids the official model pages list. OpenAI's own guidance for retiring a model is identical: replace the entry with a model available to your account and update scripts and codex exec --model references. Writing an unreleased string into model comes with no official guarantee.
The usage name and the picker disagree. What does that prove?
Only that someone saw that line in their usage. cellcog.ai (2026-11-01, community) concludes a usage label alone cannot show which model ran; OpenAI has confirmed nothing, and the official changelog and Codex docs fetched on 2026-11-02 do not contain the name.
Which id do I write when Codex runs through QCode?
Follow the Codex page on docs.qcode.cc: GPT goes over the OpenAI Responses protocol, so config.toml sets base_url to https://api.qcode.cc/openai and wire_api to responses, and model takes the listed ids gpt-6.1-sol, gpt-6.1-terra, gpt-6-sol, gpt-6-astra and gpt-5.6-terra; QCode does not support gpt-6-luna or gpt-5.6-luna for now, so please use another GPT model. What is listed is decided by /models, billing is per token, and per-model rates are on /models.
Sources
Official: the API changelog, model pages and Responses API reference on the OpenAI developer site (developers.openai.com), the Codex model, slash-command and configuration references (learn.chatgpt.com and developers.openai.com/codex), and the GitHub releases of openai/codex - all fetched 2026-11-02. Community: the X posts of @leo114119, @Mr_Salio, @Mashiro_0619, @ai_funss, @PokeBenchTV, @EthanAgentLab and @AGTPinsights, plus pages on cellcog.ai, artificialwatch.com and thewincentral.com, fetched 2026-11-02. QCode: qcode.cc/models and the Codex page on docs.qcode.cc, fetched 2026-11-02.
Run the released models first
The same QCode key switches between gpt-6.1-sol, gpt-6-sol and gpt-6-astra by changing model; for Codex write the base_url and wire_api from the docs page, billed per token, with per-model rates on /models.
Related reading
GPT-6.2 release date
Official pages give no date; that page keeps official wording and community claims apart.
Codex CLI third-party API setup
config.toml, base_url and the errors you hit while connecting.
Pay-as-you-go Packs: no subscription, 7 days
Buy one quota and spend it per token; per-model rates are on /models.
Crypto payment guide
Nine coins, 56 networks, zero platform fee for AI coding API credits.
Everything here was checked on 2026-11-02: official pages are recorded as fetched, and every community post is attributed and unconfirmed by OpenAI. Prices and availability follow the official pages of the day, and on the QCode side /models is the reference; this page will be updated when OpenAI ships.