GPT-6.2 Astra Spotted in Codex Usage — Not Released Yet, But a Friday Launch Is Reported

Is GPT-6.2 Astra out? No — not yet. On November 1, 2026, several Codex users opened their ChatGPT usage analytics and found a model OpenAI has never announced: gpt-6.2-astra, listed next to the current GPT-6 Astra with real usage attached. But try to select it and Codex says the model does not exist. A separate report says new OpenAI models could arrive as soon as Friday, November 6. The name is real — it appeared in OpenAI’s own usage data — and unconfirmed in every other sense. Here is what is confirmed, what is only reported, and what to ignore.

GPT-6.2 Astra: Confirmed vs Reported vs Not True

What Exactly Showed Up in Codex?

The story started the way most real AI leaks start: not with a press release, but with a screenshot. From November 1, Codex users posted images of their ChatGPT usage analytics — the page showing which models handled your requests — and there it was: gpt-6.2-astra, next to the familiar gpt-6-astra. Usage was dated October 31, with shares from around 1.6% of one person’s day to 100% of another’s — a pattern that looks less like a typo and more like a model briefly live for a small group, then pulled back. Independent leak-tracking sites — CellCog, OmniaKey, BeatAPI and Router/FloaTai — all reported the same screenshots and name.

But the model does not work: the picker still says GPT-6 Astra, and calling gpt-6.2-astra returns a model_not_found error. Usage analytics only list models that processed requests, so something inside OpenAI’s systems answered requests under that name on October 31 — an internal test, a switched-off rollout, a logging glitch; nobody outside OpenAI can say yet.

What OpenAI Has Actually Said: Nothing About 6.2

As of today, November 6, 2026, OpenAI has not announced any model called GPT-6.2 Astra. Its model documentation lists no 6.2 model; its API changelog runs to October 30, 2026, ending with GPT-6 Rosalind; OpenRouter has no listing, and the model’s own page returns a 404. A launched model does not hide from all four of those places at once.

The closest thing to an official hint came from Sam Altman himself. At ADIPEC in Abu Dhabi on November 2, he said the next GPT is coming soon, describing the model behind a Navier–Stokes Millennium Prize run that used 10,000 agents over about four days. But he gave no name, no version number and no date — it confirms something is coming, not that GPT-6.2 Astra is the thing.

The Friday, November 6 Report — Read It Carefully

The reason today, Friday, November 6, is circled on everyone’s calendar is a report from The Information, published November 4. It said new OpenAI models “could come as soon as Friday”, and that the release “could include the next generation of its Astra, Sol and cyber focused models”. Two words do a lot of work there: could, twice. The report added that “model timelines could change”. It is paywalled, so most people repeating it have not read the full item.

Just as important is what the report did not say. It never mentioned the name GPT-6.2 Astra. That name comes only from the Codex screenshots; the Friday timing comes only from The Information. Neither source confirms the other. Recent history also argues for caution: the Journal reported on September 28 that OpenAI scrapped GPT-6.1 Astra over safety concerns, and Polymarket put a new Astra release in the week of November 2 at about 37%.

Why “6.2” Makes Sense — and What You Can Use Today

The version number fits: the current model is GPT-6 Astra, released September 3, 2026, so a 6.2 follow-up would be a normal incremental step, especially if 6.1 was built and shelved. But leaked names can be test labels or routing tags that never become public names — the screenshots prove the string existed in OpenAI’s usage pipeline on October 31, not that OpenAI plans to ship a model under that name.

The newest Astra model you can genuinely use right now is GPT-6 Astra: a 1,050,000-token context window, up to 128,000 output tokens, a knowledge cutoff of April 30, 2026, and API pricing of $10 per million input tokens and $50 per million output tokens. That context window matters most for developers — a million-plus tokens means feeding the model an entire codebase in one session. It is also why Codex users care: Codex is where Astra models do their heaviest work, and a new version appearing in Codex usage data first is the pattern you would expect from a coding-focused successor in testing. For readers in Pakistan, nothing about your ChatGPT or Codex account changes today — a new model would appear in the picker and the changelog first, not on leak blogs.

How to Check This Yourself (and Not Get Fooled)

AI model leaks are now a weekly event, and most fall apart on the second look. Check the model where it cannot be faked easily: the official documentation, the API changelog, and the model picker. Then look for the same claim from two independent sources — here, several Codex users posting their own screenshots counted, because each screenshot was its own evidence. Finally, be suspicious of any leak that comes with precise benchmark scores or exact pricing: the easiest numbers to invent, the hardest to verify before launch. We have left every one of them out of this article on purpose.

AI is landing in everyday apps at the same time: WhatsApp is testing message scheduling, the OnePlus 16 has launched with its Pakistan price and specs, and Apple is pushing its AI hardware vision with HomeView, its new Siri-powered AI device. The GPT-6.2 Astra story sits in that same fast-moving week — which is exactly why it is worth getting right rather than getting first.

New Signals on November 6: An Unnamed Internal Model and a Brief API Flash

Friday has brought fresh signals — but still no announcement. Two separate developments are worth recording, and neither one confirms GPT-6.2 Astra.

First, OpenAI’s own research post from November 3, titled “Accelerating engineering for the physical world”, quietly revealed an unnamed internal model that comfortably beats GPT-6 Astra on a CAD engineering benchmark. On BenchCAD 2.0 — a test of rebuilding 3D parts and assemblies from drawings and reading circuit boards — the unnamed model averaged 0.875, against 0.594 for GPT-6 Astra at its highest reasoning setting. Even the new model’s cheapest setting, at 0.758, already beats Astra’s most expensive one. OpenAI also said model usage across its own hardware teams, measured in tokens, has grown 23-fold since February, and announced a new research collaboration with Mazda plus expanded work with Chip Ganassi Racing. What OpenAI did not do is name the model, connect it to GPT-6.2, or say when — or whether — anyone outside the company will ever use it.

Second, two never-before-seen model IDs — rc-ada-or-vm and rc-ada-sh-vm — briefly appeared in OpenAI’s official API model list and were then removed, according to reports from 36Kr and HTX citing posts on X. The reports note that “rc” stands for release candidate and that “ada” is the internal codename for this generation of models. OpenAI has not confirmed any of this, and a model ID flashing in a list for minutes is not a launch.

There is also one confirmed addition that is easy to misread: OpenAI added GPT-6-Cyber to its API changelog on November 5. Its system card describes it as a cyber-specialised variant of GPT-6 Astra, with initial access limited to selected approved Daybreak Red customers through the API and Codex. Nothing in the changelog identifies it as Astra 6.2.

So the position at midday on Friday, November 6, is unchanged in the one respect that matters: OpenAI has still not announced GPT-6.2 Astra. What has changed is the weight of the smoke — an unnamed model beating Astra by a wide margin in OpenAI’s own published benchmark, release-candidate IDs surfacing in the API list, and a new cyber variant quietly entering the changelog. Any one of those could be routine internal testing. Together, they are the reason the Friday report is still being watched.

More November 6 Signals: Google Cloud Quota Screenshots — Still No Announcement

Updated: Friday, November 6, 2026 (evening PKT). Two more unconfirmed sightings surfaced around the reported Friday window — and OpenAI has still announced nothing.

Screenshots of Google Cloud Console quota pages posted on November 4–5 show the strings gpt-6.2-astra and gpt-6.2-sol, according to independent tracker i10x.ai. A separate screenshot of a leaked internal tracker, shared on X on November 4, shows a model labelled gpt-6.2-betelgeuse. Neither sighting has been confirmed by OpenAI or Google, and screenshots alone cannot prove which model actually ran or when it will ship.

The Information’s reported window — new Astra, Sol and cyber-focused models “could come as soon as Friday”, November 6 — is lapsing with no OpenAI announcement, no model card, no pricing and no entry in OpenAI’s public model docs or API changelog. As of Friday evening in Pakistan, the position is unchanged: GPT-6.2 Astra has still not been announced or released. We will update this article the moment OpenAI says anything official.

What Happens Next

Friday, November 6, could bring an announcement, or nothing at all. The Information’s report gives the day a real basis, Altman’s remarks give the week a sense of imminence, and the screenshots give the name a paper trail. But a cancelled 6.1, a model_not_found error and a 37% betting market all point the other way. If OpenAI announces something, expect it in the changelog and the model picker within minutes, and we will update this article the moment it says anything official. Until then: GPT-6.2 Astra was spotted in Codex usage logs, it is not released, it cannot be used, and a Friday launch is reported but not confirmed. Everything else you read today is somebody guessing.

Frequently Asked Questions

By Fatima Raza — November 6, 2026. This article will be updated as soon as OpenAI makes an official announcement.